Researchers introduce MotifGen, a generative model designed for the spatiotemporal interpolation of tropical cyclone microwave images from multiple geospatial sources with irregular time intervals and geographic misalignment. The model addresses the challenge of high heterogeneity in microwave data by combining inputs from various instruments to fill gaps caused by long satellite revisit times.
- Trained via a self-supervised task where a random source is masked and reconstructed, leading to a significant decrease in Continuous Ranked Probability Score compared to supervised training.
- Combining infrared and microwave data yields further performance improvements over using microwave data alone.
- The model produces an ensemble mean comparable to deterministic models while generating a power spectrum significantly closer to true observations.
This approach provides the first generative model capable of interpolating cyclone microwave images by integrating multiple microwave instruments and infrared observations at irregular intervals.