Predictive Statistical Representations of Observed and Simulated Rainfall Using Generalized Linear Models.

Area of Science:

• Atmospheric Science
• Climate Modeling
• Remote Sensing

Background:

• Tropical rainfall exhibits complex subdaily variations and spatial patterns.
• Accurate prediction of rainfall is crucial for understanding climate dynamics and impacts.
• Existing climate models may have limitations in simulating convective processes.

Purpose of the Study:

• To assess the feasibility of predicting tropical Pacific rainfall occurrence and amount from atmospheric profiles.
• To differentiate prediction capabilities across various rain types (stratiform, deep convective, shallow convective).
• To compare statistical model performance trained on satellite observations versus climate model simulations.

Main Methods:

• Generalized linear models (logistic and gamma regression) were employed.
• TRMM satellite radar data and CAM5 simulations were used for rainfall information.
• MERRA-2 and CAM5 environmental variables served as predictors.
• Models were trained on 2003 data to predict 2004 rainfall.

Main Results:

• Humidity and temperature profiles were key predictors for rainfall.
• Logistic regression performed well for rain occurrence, particularly in the east Pacific.
• Gamma regression provided reasonable geographical rain amount distributions but struggled with rain rate probabilities.
• Models trained on observations indicated potential deficiencies in CAM5's convection parameterization.

Conclusions:

• Statistical models using satellite observations and reanalysis data can predict time-averaged tropical rainfall patterns.
• The study highlights discrepancies between observed and simulated rainfall, suggesting areas for climate model improvement.
• Further development of models is needed for accurate prediction of rain rates.