Traditional deterministic and geostatistical methods for rainfall interpolation usually fall short of integration of data on a variety of variables. These omitted variables include seasonal variables such as time of year, topographic variables such as elevation, and/or remote sensing variables such as radar reflectivity. Meanwhile, poor quality in data on certain variables for some data points poses challenges to modelers who are using machine learning approaches to estimate rainfall amounts for locations without gauge measurements. To overcome these limitations, this presentation introduces a novel deep learning-based approach to recreate rainfall histories for large geographic areas with a high spatio-temporal resolution. The proposed approach enables integration of data on a variety of variables by adopting a multi-layer perceptron modeling framework. The introduction of binary variables on data quality as additional input variables resolves the issue of unequal data quality for different data points. As a demonstration, historical records of rainfall at hourly and daily intervals recorded at 139 rain gauge stations in or close to Harris County, Texas, from 1986 to 2013 are used, along with other auxiliary variables, to train deep learning regression models to interpolate rainfall at surface level. Results of validation and recreated spatiotemporal distributions of rainfall indicate good performance of the proposed approach compared to both gauged and radar data. The final product of the proposed approach can be applied to other regions, with information on hindcast historical rainfall events, for pluvial flood risk analysis. The approach will assist researchers and policy specialists to validate hydrologic modeling as well as for training machine learning models to identify extreme rainfall events to facilitate early warning and emergency response.

We present an interdisciplinary study of observations of pre-earthquake processes associated with major earthquakes based on integrating space and ground- data. Recent large magnitude earthquakes in Asia and Europe have emphasized the various observations of multiple types of pre-earthquake signals recorded either on the ground or from space. Four physical parameters were measured from ground and satellite and used in our simulation models: 1) Ground Radon variation; 2) Outgoing Long-Wavelength Radiation (OLR) obtained from NPOES, NASA/AQUA) on top of the atmosphere (TOA); 3) Atmospheric Chemical Potential (ACP) obtained from NASA assimilation models and; 4) electron density variations in the ionosphere via GPS Total Electron Content (GPS/TEC). For this analysis we selected six large earthquakes from the last decade with differing geographic and seismo-tectonics regions: (1) M9.3, Off the West Coast of Northern Sumatra, Dec 26, 2004; (2) M9.0 Great Tohoku Earthquake, Japan, March 11, 2011; (3) and (4) M7.8 and M7.3 Gorkha, Nepal, 2015; (5); M8.2 Tehuantepec, Mexico, September 8, 2017 and; (6) M7.1, Puebla central Mexico earthquakes, September 19, 2017. Our preliminary results indicate an enhancements of radon (about a week to ten days prior) coincident (with some delay) with an increase in the atmospheric chemical potential measured near the epicenter from both satellite and subsequently with an increase of outgoing infrared radiation (OLR) observed on the TOA from NOAA/NASA (a week in advance). Finally GPS/TEC data indicate an increase of electron concentration 1-4 days before the earthquakes. Although the radon variations and some of satellite OLR anomalies were observed far (>2000km) from the epicenter areas the anomalies were always inside the estimates of the Dobrovolsky-Bowman area of preparation. We examined the possible correlation between magnitude and the spatial size of earthquake preparation zone in the framework of the Lithosphere –Atmosphere -Ionosphere Coupling hypothesis. The reliable detection of pre-earthquake signals for both sea and land earthquakes was possible only by integrating satellite and ground observations. A detail summary of our approach to this study of pre-earthquake research has just been published as AGU/Wiley Geophysical Monograph Series No. 234.