A Machine Learning Perspective On Improving Numerical Weather Prediction

Abstract

Accurate weather forecasting is crucial for agri- culture, disaster preparedness, and daily planning. Traditional Numerical
Weather Prediction (NWP) models, which depend on complex physical equations, often struggle with inaccuracies due to uncertainties in
initial conditions and high computational demands. Machine Learning (ML) techniques offer a powerful alternative by leveraging historical
data patterns to predict future weather events more effectively.
This study explores the implementation of six ML mod- els Naıve Bayes, Support Vector Machine (SVM), Logistic Regression, Decision Tree,
Random Forest, and Gradient Boost- ing for short-term weather prediction. These models were trained on historical meteorological data and
evaluated based on accuracy, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). Results indicate that ensemble-based
models, particularly Random Forest and Gradient Boosting, outperform individual classifiers by improving accuracy and minimizing
prediction errors.
This research highlights the potential of ML in modern weather forecasting and discusses future advancements, including deep learning
applications and real-time data integration, to enhance forecasting accuracy and reliability.