We studied disastrous heavy rainfall episodes in 2017–2019 summer in SW Japan, especially in the Kyushu region using tropospheric delay data from the Japanese dense global navigation satellite system (GNSS) network GEONET (GNSS Earth Observation Network). This region often suffers from extremely heavy rains associated with stationary fronts during summer. In this study, we first analyze behaviors of water vapor on July 6, 2018, using tropospheric parameters obtained from the database at the University of Nevada, Reno. The data set includes tropospheric delay gradient vectors (G), as well as zenith tropospheric delays (ZTD), estimated every 5 min. At first, we interpolated G to obtain those at grid points and calculated their convergence, similar to the quantity proposed by Shoji (2013) as water vapor concentration (WVC) index. We obtained zenith wet delay (ZWD) from ZTD by removing zenith hydrostatic delay. The raw ZWD values do not really reflect the wetness of the atmosphere above the GNSS station because they largely depend on the station altitudes. To study the dynamics of water vapor before heavy rains, we estimated ZWD converted to the values at sea level. In the inversion scheme, we used G at all GEONET stations and ZWD data at low-altitude (<100 m) GEONET stations as the input. Then we assumed that spatial change of ZWD is proportional to G (e.g., G_x = H ∂ZWD/∂x, where H is the water vapor scale height) and estimated sea-level ZWD at grid points all over Japan. At last, we tried to justify our working hypothesis that heavy rain occurs when both WVC and sea-level ZWD are high by analyzing hourly water vapor distributions in all the days in July 2017, July 2018, and August 2019. We found that both two values showed maxima in the studied period when the three heavy rainfall (>50 mm/h) episodes occurred, that is, July 5, 2017, July 6, 2018, and August 27, 2019. Next, we performed high time resolution analysis (every 5 min) on the days of heavy rain. The results suggest that both WVC and sea-level ZWD go up prior to the onset of the rain, and ZWD decreases rapidly once the heavy rain started. It is a future issue, however, how far these two quantities contribute to forecast heavy rains.

我们使用来自日本密集全球导航卫星系统（GNSS）网络GEONET（GNSS地球观测网络）的对流层延迟数据，研究了日本西南部2017-2019年夏季灾难性的强降雨事件，特别是在九州地区。在夏季，该地区经常遭受与固定锋面相关的特大暴雨。在这项研究中，我们首先使用从内华达大学里诺分校的数据库获得的对流层参数，于2018年7月6日分析水蒸气的行为。数据集包括对流层延迟梯度向量（G），以及天顶对流层延迟（ZTD），每5分钟估算一次。首先，我们插值G 以获得网格点上的那些并计算它们的收敛，类似于 Shoji（2013）作为水蒸气浓度（WVC）指数。通过消除天顶静水延迟，我们从ZTD获得了天顶湿延迟（ZWD）。原始ZWD值并不能真正反映GNSS站上方大气的湿度，因为它们很大程度上取决于站的高度。为了研究大雨前的水汽动力学，我们估算了ZWD转换为海平面上的值。在反演方案中，我们使用了G在所有GEONET站输入，并在低空（<100 m）GEONET站输入ZWD数据。然后我们假设ZWD的空间变化与G （例如， G_X = H WDZWD /∂X，在哪里 H是整个日本的网格点的水蒸气标高）和估计的海平面ZWD。最后，我们通过分析2017年7月，2018年7月和2019年8月全天的每小时水蒸气分布来证明当WVC和海平面ZWD都很高时会出现大雨的工作假设是正确的。我们发现，这两个值显示了在研究期间发生的三场强降雨（> 50 mm / h）的最大值，分别是2017年7月5日，2018年7月6日和2019年8月27日。接下来，我们进行了高时间分辨率分析（每5分钟一次）。结果表明，在降雨开始之前WVC和海平面ZWD都会升高，一旦大雨开始，ZWD会迅速下降。但是，这两个问题对预测暴雨有多大作用，这是一个未来的问题。