Jay Uday Potnis
M.S. FinalMay 10, 2022, 9:00 am - 10:15 am
Teams
DEEP NEURAL NETWORK BASED RADAR RAINFALL CLASSIFICATION AND ESTIMATION
Abstract: Quantitative Precipitation Estimation is the process of computing rainfall rate or rainfall accumulation
based on the state of the atmosphere. Atmospheric conditions can be described by using
observations from meteorological instruments. Extreme weather events caused due to high rainfall
can be dangerous in terms of loss of property and life. To prevent such disasters, accurate QPE
algorithms that analyze and estimate the amount of rainfall observed in a region are critical. Moreover,
rain rate estimates are crucial products in making management decisions in water, energy,
construction infrastructure, and many other institutions. Researching state-of-the-art rainfall estimation
techniques that make use of reliable remote sensing equipment such as satellites and radars
is important as deploying rain gauges everywhere is not possible and is not a viable option. As
rain precipitation is a complicated phenomenon, depending on multiple factors in the atmosphere,
research is being done in this domain for many decades and the goal is to improve the accuracy of
estimation by using new state-of-the-art methods.
Weather radars are reliable remote sensing instruments that are used to capture the different
properties of weather in form of products called moments. The goal of this work is to use weather
radars in conjunction with Deep Neural Networks to provide solutions to multiple tasks in the
QPE domain. Neural networks can be used for precipitation flagging such as classifying rain and
no rain events. They can also be used for estimating the rain rates at specific coordinates or along
regions. Though multiple empirical relationships between radar moments and rain rate already
exist, this work provides good state-of-the-art alternatives to these equations and can even achieve
comparable accuracy.
based on the state of the atmosphere. Atmospheric conditions can be described by using
observations from meteorological instruments. Extreme weather events caused due to high rainfall
can be dangerous in terms of loss of property and life. To prevent such disasters, accurate QPE
algorithms that analyze and estimate the amount of rainfall observed in a region are critical. Moreover,
rain rate estimates are crucial products in making management decisions in water, energy,
construction infrastructure, and many other institutions. Researching state-of-the-art rainfall estimation
techniques that make use of reliable remote sensing equipment such as satellites and radars
is important as deploying rain gauges everywhere is not possible and is not a viable option. As
rain precipitation is a complicated phenomenon, depending on multiple factors in the atmosphere,
research is being done in this domain for many decades and the goal is to improve the accuracy of
estimation by using new state-of-the-art methods.
Weather radars are reliable remote sensing instruments that are used to capture the different
properties of weather in form of products called moments. The goal of this work is to use weather
radars in conjunction with Deep Neural Networks to provide solutions to multiple tasks in the
QPE domain. Neural networks can be used for precipitation flagging such as classifying rain and
no rain events. They can also be used for estimating the rain rates at specific coordinates or along
regions. Though multiple empirical relationships between radar moments and rain rate already
exist, this work provides good state-of-the-art alternatives to these equations and can even achieve
comparable accuracy.
Adviser: V. Chandrasekar
Co-Adviser: N/A
Non-ECE Member: Thomas Siller
Member 3: Margaret Cheney
Addional Members: N/A
Co-Adviser: N/A
Non-ECE Member: Thomas Siller
Member 3: Margaret Cheney
Addional Members: N/A
Publications:
Jay Potnis, V.Chandrasekar, & Ryan Gooch. (2021). Real Time Alignment and Distribution of Weather Radar Data with Rain Gauge Data for Deep Learning using CHORDS (v1.0). Zenodo. https://doi.org/10.5281/zenodo.5496308
Jay Potnis, V.Chandrasekar, & Ryan Gooch. (2021). Real Time Alignment and Distribution of Weather Radar Data with Rain Gauge Data for Deep Learning using CHORDS (v1.0). Zenodo. https://doi.org/10.5281/zenodo.5496308
Program of Study:
ECE 528
CS 545
CS 530
ECE 554
ECE 571
ECE 575
ECE 561
ECE 450
ECE 528
CS 545
CS 530
ECE 554
ECE 571
ECE 575
ECE 561
ECE 450