Improving multiple model ensemble predictions of daily precipitation and temperature through machine learning...

Area of Science:

• Climate Science
• Machine Learning Applications
• Hydrological Modeling

Background:

• General Circulation Models (GCMs) are crucial for climate simulations but often require improvement.
• Multi-Model Ensembles (MMEs) are a common technique to enhance GCM performance.
• Evaluating various MME techniques is essential for accurate climate projections.

Purpose of the Study:

• To assess the efficacy of different MME techniques for precipitation and temperature.
• To compare traditional averaging methods with advanced machine learning approaches.
• To identify optimal MME strategies for a tropical Indian river basin.

Main Methods:

• Utilized 21 GCMs from NASA NEX-GDDP and 13 GCMs from CMIP6 datasets.
• Applied arithmetic mean, Multiple Linear Regression (MLR), Support Vector Machine (SVM), Extra Tree Regressor (ETR), Random Forest (RF), and Long Short-Term Memory (LSTM) for ensembling.
• Evaluated MME performance using coefficient of determination (R²).

Main Results:

• LSTM demonstrated superior performance for precipitation MMEs, achieving an R² of 0.9.
• All machine learning methods outperformed the arithmetic mean ensemble.
• RF and LSTM showed consistent high performance for temperature MMEs, with R² values from 0.82 to 0.93.

Conclusions:

• Recommends RF and LSTM methods for developing MMEs in the studied basin.
• Machine learning approaches offer significant advantages over the mean ensemble method for climate data.
• Advanced MME techniques are vital for improving the accuracy of climate projections.