Intermediate Weather Data Generation with an Intermediate AutoEncoder
Oct - Dec 2024
Yuzhang Huang*, Rico Rodriguez Passanisi*, Devin Sze*
Weather data is often segmented into specific time intervals, which limits the granularity of the data to some arbitrarily large time interval. Using Intermediate Autoencoders (IAEs), this model is able to successfully generate intermediate weather data between recorded timestamps. The IAE model is trained on a dataset of fused images, each comprising three sub-images representing the "before", "normal noise", and "after" weather conditions. These images included areas with significant precipitation, areas without any, and a variety of weather patterns in between. While there were edge cases that the model did not perform well on, the IAE model was an effective predictor in the majority of precipitation cases. By effectively predicting intermediate weather patterns, this model has applications in real-time weather forecasting and analysis, but can also be applied to other applications like medical symptom progression and animation frame generation.
*Equal Contributor
