Scholarlink

<p>Regional modeling of zenith tropospheric delay (ZTD) is fundamental for high-precision Global Navigation Satellite System (GNSS) positioning and atmospheric research, yet capturing complex, nonlinear atmospheric variability remains challenging. This study presents a novel framework for regional ZTD prediction by integrating spatially constrained multivariate clustering (SCMC) with GBM, LightGBM, and XGBoost. The proposed framework aims to improve prediction accuracy for regional atmospheric structures exhibiting spatiotemporal variability. Using 43 GNSS stations, five spatial clusters were identified across the study area using the SCMC algorithm based on ECMWF Reanalysis v5 (ERA5) meteorological variables. The cluster-specific prediction models were trained on Vienna Mapping Functions 3 (VMF3) features [2020–2023] and tested using 2024 data, with the IGS final tropospheric product (IGS<sub>ZPD</sub>) and the ECMWF Integrated Forecast System (ECMWF<sub>IFS</sub>) serving as benchmarks to validate the accuracy and generalizability of the regional models. Results indicate that spatial clustering enhances predictive accuracy and regional adaptability. Among the tested models, XGBoost exhibited the most substantial improvement; while the model trained over the entire study area yielded a root mean square error (RMSE) of 2.0&#xa0;cm, cluster-specific models achieved RMSE values as low as 1.1&#xa0;cm, representing an impressive 35% accuracy enhancement. Consistent gains of approximately 19 and 16% were also observed for LightGBM and GBM, respectively. These findings demonstrate that the SCMC-integrated XGBoost model effectively captures ZTD variations under complex climatic conditions, exhibiting superior predictive capabilities. Furthermore, the computational efficiency and low-latency inference of LightGBM and XGBoost make them ideal for real-time GNSS applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel regional tropospheric delay prediction model based on spatial clustering and gradient boosting algorithms

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月22日

查看全文
  • Ali Utku Akar,
  • Cevat Inal

摘要

Regional modeling of zenith tropospheric delay (ZTD) is fundamental for high-precision Global Navigation Satellite System (GNSS) positioning and atmospheric research, yet capturing complex, nonlinear atmospheric variability remains challenging. This study presents a novel framework for regional ZTD prediction by integrating spatially constrained multivariate clustering (SCMC) with GBM, LightGBM, and XGBoost. The proposed framework aims to improve prediction accuracy for regional atmospheric structures exhibiting spatiotemporal variability. Using 43 GNSS stations, five spatial clusters were identified across the study area using the SCMC algorithm based on ECMWF Reanalysis v5 (ERA5) meteorological variables. The cluster-specific prediction models were trained on Vienna Mapping Functions 3 (VMF3) features [2020–2023] and tested using 2024 data, with the IGS final tropospheric product (IGSZPD) and the ECMWF Integrated Forecast System (ECMWFIFS) serving as benchmarks to validate the accuracy and generalizability of the regional models. Results indicate that spatial clustering enhances predictive accuracy and regional adaptability. Among the tested models, XGBoost exhibited the most substantial improvement; while the model trained over the entire study area yielded a root mean square error (RMSE) of 2.0 cm, cluster-specific models achieved RMSE values as low as 1.1 cm, representing an impressive 35% accuracy enhancement. Consistent gains of approximately 19 and 16% were also observed for LightGBM and GBM, respectively. These findings demonstrate that the SCMC-integrated XGBoost model effectively captures ZTD variations under complex climatic conditions, exhibiting superior predictive capabilities. Furthermore, the computational efficiency and low-latency inference of LightGBM and XGBoost make them ideal for real-time GNSS applications.