Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off
One annual plan covers every course and certificate on Coursera. 40% off for a limited time.
Get Full Access
Learn how to develop neural network surrogate models for atmospheric cloud microphysics through a progressive refinement sampling strategy that mimics maximal information entropy distribution. Explore the challenges of training machine learning models on atmospheric data where over 99.9% of values represent quiescent clear air conditions with near-zero derivatives for potential temperature, specific humidity, and cloud water content variables. Discover how traditional filtering approaches create imbalanced datasets that over-sample uninteresting near-zero values while examining a novel sampling strategy that progressively trains on larger data samples chosen to flatten the phase-space distribution. Understand the mathematical foundations behind information-entropy-maximizing distributions and their application to accelerating convergence toward accurate neural network surrogate models for the Intermediate Complexity Atmospheric Research (ICAR) model. Gain insights into handling probability distributions that span nine decades of values and learn practical approaches for managing computational costs while maintaining model accuracy in earth system simulation applications.
