The study significantly reduced the estimated errors and uncertainties of the rain rate by combining observed raindrop size and radar reflectivity using machine learning.
Flooding from rainfall is one of the most damaging and deadly hazards associated with tropical cyclones, so estimating rainfall amounts is important to keep people and property safe. Forecasters use weather radar data to estimate how much rain is falling based on the complex relationship between what the radar measures, known as radar reflectivity, and the amount of rain that falls during a period of time, known as the rain rate. This relationship changes depending on the type of rain (such as drizzle, snow, or strong thunderstorms). Different relationships have been created depending on where in the tropical cyclone the rainfall is happening (like the eyewall or outer rainbands) or the tropical cyclone’s development stage (for example, whether it is just getting organized, rapidly intensifying, or decaying). This study used machine learning on radar reflectivity measurements and data from probes that measure raindrop size aboard NOAA Hurricane Hunter aircraft to reduce errors in estimating rainfall from radar reflectivity measurements. In addition to estimating rainfall during landfall, the new relationship can be used to improve the way rainfall is included in computer models.

Important Conclusions:
- Errors and uncertainties in estimating rainfall amounts from reflectivity are mainly associated with the size of the raindrops.
- Errors and uncertainties in estimating rainfall amounts are greatly reduced using a new machine learning tool.

For more information, contact aoml.communications. The paper can be accessed at https://agupubs.onlinelibrary.wiley.com/doi/epdf/10.1029/2022GL099332