Presentation Title:A Hybrid Machine Learning Framework for Enhanced Precipitation Nowcasting
Full Name:Luca Delle Monache
Affiliation / Institution:CW3E/SIO/UCSD
Co-Author
Full Name:Duncan Axisa
Affiliation / Institution:CW3E/SIO/UCSD
Co-Author 2
Full Name:Vesta Afzali Gorooh
Affiliation / Institution:CW3E/SIO/UCSD
Co-Author 3
Full Name:Zhenhai Zhang
Affiliation / Institution:CW3E/SIO/UCSD
Co-Author 4
Full Name:Mohammadvaghef Ghazvinian
Affiliation / Institution:CW3E/SIO/UCSD
Co-Author 5
Full Name:Chandrasekaran Venkatachalam
Affiliation / Institution:Colorado State University
Co-Author 6
Full Name:Kim Eun Yeol
Affiliation / Institution:Colorado State University
Co-Author 7
Full Name:Chandrasekar Radhakrishnan
Affiliation / Institution:Colorado State University
Co-Author 8
Full Name:Ernesto Damiani
Affiliation / Institution:Khalifa University
Co-Author 9
Full Name:Marco Anisetti
Affiliation / Institution:Khalifa University
Abstract:Operational rainfall enhancement programs rely heavily on the accuracy of precipitation forecasts to improve cloud targeting, which guides the action of inserting seeding material at the right place and at the right time in the vicinity of suitable clouds. As part of the UAE Research Program for Rain Enhancement Science (UAEREP) we are developing artificial intelligence (AI) algorithms to improve cloud seeding operations over the UAE. We have built a novel AI framework leveraging satellite observations, ground-based weather radar data, rain gauges, aircraft data, and numerical weather prediction estimates to extract features and generate products to determine optimal cloud seeding timing and location, and to generate more accurate quantitative precipitation estimation to support the UAE rainfall enhancement program. An advanced deep learning algorithm is proposed to learn from several thousands of examples from historical data how to effectively extract and extrapolate inputs and the required cloud features important to define cloud patches that are seedable. These features and inputs, along with extrapolated satellite and radar data, as well as numerical weather prediction data and rain gauges, are utilized as input to an AI-based model to generate precipitation predictions six hours in the future. We will describe the methods developed under this project to support cloud seeding operations and provide an assessment of their performance.