Rainfall Estimation from TEMPEST-D CubeSat Observations: A Machine Learning Approach

The Temporal Experiment for Storms and Tropical Systems (TEMPEST) is a 6U CubeSat mission to measure the temporal evolution of clouds, convective systems and the surrounding water vapor environment using global observations with resolution of several minutes. The TEMPEST constellation mission consists of 6-8 identical 6U CubeSats deployed in the same orbital plane with 3-4 minute spacing. The necessary technology for the success of the TEMPEST constellation mission has been demonstrated during the TEMPEST-D ("D" for demonstration) satellite mission launched on May 21, 2018 and deployed into orbit on July 13, 2018. The TEMPEST-D CubeSat radiometers measure at five millimeter-wave frequencies (87, 164, 174, 178 and 181 GHz) to demonstrate the capability to provide precise observations of convection and vertical profiles of the surrounding water vapor.

At present, quality-controlled TEMPEST-D observations are available from September 2018 to July 2019. A significant number of storm events are captured by TEMPEST-D over the Continental United States (CONUS). The observations show that the pattern of storms observed by TEMPEST-D brightness temperatures (TBs) exactly matches the ground-based radar Quantitative Precipitation Estimation (QPE) data in space and time. Multi-Radar and Multi-Sensor ( MRMS) QPE data from the NOAA National Severe Storms Laboratory (NSSL) is used as ground truth. Figure 1 shows the TEMPEST-D TBs (in K) at the five millimeter-wave frequencies and MRMS QPE (in mm/hr) data over a storm over the U.S. Midwest and Great Plains on May 19, 2019. The figure shows the apparent spatial correlation between the space-borne TEMPEST-D radiometer measurements and ground-based radar observations. The initial quantitative comparison highlights the capability and quality of TEMPEST-D data and provides motivation for estimating rainfall from TEMPEST-D observations.

The objective of this research work is to estimate the rainfall on the ground from TEMPEST-D TB measurements using a machine learning approach. This study will focus on developing a machine learning based algorithm to estimate rainfall from TEMPEST-D measurements. The MRMS QPE data will be considered as truth in the data training as well as independent validation of estimated rainfall from the machine learning algorithm. This study will also analyze and compare the rainfall estimated from TEMPEST-D measurements with that estimated from the Global Precipitation Measurement (GPM) mission. The findings and detailed results of this evaluation will be presented at the conference.