Accurate forecasting of short-term convective heavy precipitation remains a significant challenge due to the high spatiotemporal variability of atmospheric water vapor. Global Navigation Satellite System (GNSS) Zenith Total Delay (ZTD) can provide valuable constraints on the atmospheric moisture field for Numerical Weather Prediction (NWP) models. However, traditional geodetic-grade GNSS stations are often spatially sparse. This study investigates the potential of assimilating GNSS ZTD observations from a high-density "opportunistic" network of the 5^th Generation Communication (“5G”) cellular base stations to improve convective forecasting over Guangdong Province, China, using a three-dimensional variational (3DVar) assimilation system. GNSS ZTD observations from more than 1,500 5G base stations were processed through a multi-stage quality-control and noise-based ranking procedure to mitigate data uncertainty. Three assimilation experiments (EXP1 to EXP3) were designed to evaluate the impact of dense GNSS ZTD assimilation. EXP1 compared the assimilation of dense GNSS ZTD with that of geodetic GNSS ZTD from the China Meteorological Administration (CMA) in a severe convective event. Overall forecast improvements were observed for both datasets. CMA GNSS ZTD performed better for light-to-moderate rainfall, whereas dense GNSS ZTD performed better for heavy rainfall. In the thinning sensitivity experiment, EXP2 showed that a thinning distance of 20 km achieved the best balance in forecast improvement, whereas denser observations with smaller thinning distances of 10-15 km degraded forecast performance, likely due to spatial error correlations and excessive local increments. EXP3 confirmed through one-month verification that the optimal 20 km thinning scheme improved both the initial moisture field and subsequent heavy rainfall forecasts (≥30 mm/12 h), with clear improvements in POD, CSI, and ETS relative to the control experiment. These results demonstrate that assimilating dense GNSS ZTD observations from 5G base stations has the potential to improve short-range precipitation forecasts and provide an effective supplement to existing terrestrial water vapor observations.
 

3-D variational (3DVar) data assimilation,,Low-cost GNSS station,Zenith total delay (ZTD),Heavy rainfall forecast