ZWDX: a global zenith wet delay forecasting model using XGBoost
OPEN ACCESS
Loading...
Date
2024-12-18
Publication Type
Journal Article
ETH Bibliography
yes
Citations
Altmetric
OPEN ACCESS
Data
Rights / License
Abstract
Tropospheric delays play a crucial role for Global Navigation Satellite Systems (GNSS). They are a major error source in GNSS positioning and, at the same time, also a variable of interest in GNSS meteorology. Regardless of whether the delay shall be eliminated or inverted to atmospheric parameters, and no matter how this is done, it is of utmost importance to accurately determine tropospheric delays. In this study, we present a global zenith wet delay (ZWD) model, called ZWDX, that offers accurate spatial and temporal ZWD predictions at any desired location on Earth. ZWDX is based on the XGBoost algorithm and uses ZWDs measured at over 19,000 GNSS stations as reference. The inputs of ZWDX are the geographical location, observation time, and specific humidity at nine atmospheric pressure levels. For our study, we train the model on the years 2010 to 2021 and then test it for the year 2022. While ZWDX is trained to predict ZWD values based on specific humidity values from the ERA5 reanalysis, we show that it also delivers good predictions when applied to HRES specific humidity forecasts, making it suitable for (short-term) ZWD forecasting. The ZWDX model predictions are evaluated at 2500 globally distributed, spatio-temporally independent GNSS stations, with forecasting horizons ranging from 0 h to 48 h, and achieve root mean squared errors (RMSE) between 10.1 mm and 16.2 mm. To independently evaluate ZWDX’s performance and to demonstrate its potential for a real-world downstream task, we use its predictions as a-priori values for a precise point positioning (PPP) analysis and compare the results with those obtained using ZWD values from VMF1 or VMF3. We find that the highest accuracy and fastest convergence are indeed achieved with ZWDX.
Permanent link
Publication status
published
External links
Editor
Book title
Journal / series
Volume
76 (1)
Pages / Article No.
163
Publisher
SpringerOpen
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Zenith wet delay (ZWD); Global predictions; XGBoost; Machine learning (ML); GNSS; Precise point positioning (PPP)
Organisational unit
03886 - Schindler, Konrad / Schindler, Konrad
09707 - Soja, Benedikt / Soja, Benedikt