The Gravity Recovery And Climate Experiment (GRACE) and GRACE-FollowOn (GRACE(−FO)) satellites have been monitoring Earth’s changes in terrestrial water storage (TWS) or surficial mass changes at monthly sampling and a spatial scale longer than ∼330 km (half wavelength) over the past two decades. At monthly sampling or revisit time, the use of satellite gravimetry is difficult to effectively monitor abrupt extreme weather events which are high-frequency, including the climate-induced hurricanes/cyclones, flash floods and droughts. The majority of the contemporary studies have focused on satellite gravimetry spatial downscaling, and not on reducing the temporal resolution of Earth’s mass change. Here, we developed a Deep Learning (DL) algorithm to downscale monthly GRACE/GRACE(−FO) Mass Concentration (Mascon) TWS anomaly (TWSA) solutions to daily sampling over the Contiguous United States (CONUS), with the aim of monitoring rapidly evolving natural hazard episodes. The simulative performance of the DL algorithm is validated by comparing the modeling to an independent observation and the land hydrology model (LHM) predicted TWSA. To confirm that our daily and monthly simulations captured the climatic variations, we first compared our simulations with El Niño/La Niña Southern Oscillation (ENSO) circulation system index, which has a dominant climatological and socioeconomic impact across CONUS, and results reveal high correlations which are statistically significant. Next, we assessed the feasibilities to detect long- and short-term variations in the TWSA signals triggered by hydrological extremes, including the 2011 and 2019 Missouri River Floods, the August 2017 Atlantic Hurricane Harvey landfalls in Texas, the 2012–2017 drought in California, and the flash drought in the Northern Great Plains in 2017. Additional validation results using independent in situ observations reveal that our DL-aided gravimetry downscaled daily simulations are capable of elucidating hazards and water cycle evolutions at high temporal resolution.

中文翻译：

---

深度学习辅助对美国本土 GRACE 衍生的陆地储水异常进行时间缩小


在过去的二十年里，重力恢复和气候实验 （GRACE） 和 GRACE-FollowOn （GRACE（−FO）） 卫星一直在监测地球的陆地储水量 （TWS） 或表面质量变化，每月采样一次，空间尺度超过 ∼330 公里（半波长）。在每月采样或重访时，使用卫星重力测量很难有效监测高频突发的极端天气事件，包括气候引起的飓风/气旋、山洪暴发和干旱。大多数当代研究都集中在卫星重力测量空间缩小上，而不是降低地球质量变化的时间分辨率上。在这里，我们开发了一种深度学习 （DL） 算法，将每月 GRACE/GRACE（−FO） 质量浓度 （Mascon） TWS 异常 （TWSA） 解决方案缩小为美国本土 （CONUS） 的每日采样，目的是监测快速演变的自然灾害事件。通过将模型与独立观测和陆地水文模型 （LHM） 预测的 TWSA 进行比较，验证了 DL 算法的模拟性能。为了确认我们的每日和每月模拟捕捉了气候变化，我们首先将我们的模拟与厄尔尼诺/拉尼娜南方涛动 （ENSO） 环流系统指数进行了比较，该指数对整个 CONUS 具有主导的气候和社会经济影响，结果揭示了具有统计意义的高度相关性。 接下来，我们评估了检测由极端水文事件触发的 TWSA 信号长期和短期变化的可行性，包括 2011 年和 2019 年密苏里河洪水、2017 年 8 月大西洋飓风哈维登陆德克萨斯州、2012-2017 年加利福尼亚州干旱以及 2017 年北部大平原的闪电干旱。使用独立原位观察的额外验证结果表明，我们的 DL 辅助重力测量缩小日常模拟能够在高时间分辨率下阐明危害和水循环演变。