Dynamic machine learning applied to merged GEO-IR and LEO-microwave data to improve the analysis and nowcasting of severe convection - HazMap4D

• Grant number: 20/11671-7
• Support Opportunities: Scholarships abroad - Research
• Effective date (Start): September 01, 2021
• Effective date (End): October 31, 2021
• Field of knowledge: Physical Sciences and Mathematics - Geosciences - Meteorology
• Principal Investigator: Daniel Alejandro Vila
• Grantee: Daniel Alejandro Vila
• Host Investigator: Francisco J. Tapiador
• Host Institution: Instituto Nacional de Pesquisas Espaciais (INPE). Ministério da Ciência, Tecnologia e Inovação (Brasil). São José dos Campos , SP, Brazil
• Research place: Universidad de Castilla-La Mancha, Toledo (UCLM), Spain

Abstract

For the monitoring of severe weather in the tropics in real time, geostationary satellites provide the best estimates of cloud top temperatures (IR) with high spatial and temporal resolution, without any specific information on the actual development under the tops of these clouds. Microwave wavelength (MW) radiometers on board low-orbit (LEO) satellites have more specific information, but only at the shortest wavelengths can the deeper convection be spatially resolved due to the pixel size at these frequencies. New methodologies recently developed at the Jet Propulsion Laboratory (JPL), which combine high-frequency passive microwave data with satellite radar data (TRMM/PR and GPM/DPR), provide specific information about the water content profile in different layers of the clouds, which makes this information a fundamental piece for the knowledge of the evolution of the deep convection on the tropics.The main objective of this project is, in a first step, to combine storms life cycle data observed in thermal infrared on geostationary satellites with data of cloud height, depth and water content above a reference level, obtained at from low orbit satellites with microwave sensors and, in a second step, produce, for a given storm, a very short-term forecast combining the available information (IR + MW) using machine learning approaches. (AU)