Statistical and Machine Learning Methods Applied to the Prediction of Different Tropical Rainfall Types

Area of Science:

• Atmospheric Science and Meteorology
• Climate Modeling
• Machine Learning Applications in Earth Science

Background:

• Predicting rainfall from large-scale environmental variables is a significant challenge for climate models.
• The efficacy of numerical methods in capturing true rainfall characteristics without storm-scale information remains uncertain.
• Rainfall exhibits diverse types (deep convective, stratiform, shallow convective) with distinct structures influencing predictability.

Purpose of the Study:

• To evaluate the predictive capabilities of three statistical and machine learning methods for rainfall occurrence and intensity.
• To assess these methods using Global Precipitation Measurement (GPM) satellite radar data and MERRA-2 reanalysis environmental profiles.
• To investigate if machine learning methods outperform traditional statistical models in predicting different rain types.

Main Methods:

• Employed three prediction methods: a generalized linear model (statistical), a neural network, and a random forest (machine learning).
• Utilized 3-hourly rain observations from the GPM satellite radar over the tropical Pacific.
• Incorporated large-scale environmental profiles (temperature, moisture) from the MERRA-2 reanalysis dataset.

Main Results:

• No single method significantly outperformed the others in predicting rain occurrence and intensity.
• All tested methods exhibited common climate model issues: over-predicting rain frequency and failing to capture extreme rain rates.
• The performance differences between statistical and machine learning approaches were not pronounced for this rainfall prediction task.

Conclusions:

• Machine learning tools require careful validation and are not universally superior for all big data challenges.
• Traditional climate modeling approaches are insufficient for accurately predicting extreme rainfall events.
• Further research into alternative methods is necessary to improve the prediction of extreme weather phenomena.