Nowcasting

Cloud and visibility nowcasting is important for operational users. These nowcasts provide end-users with critical information on near-future (0-3 hour) states of visibility, which in turn can be used for solar energy management, aviation safety, and allocation of assets. OVERCAST nowcasting efforts focus on developing novel methodologies for nowcasting 3D cloud products. These methods include extrapolation of features forward with optical flow and machine learning. Machine learning methods include using advanced convolutional neural network image-to-image translation architectures with pre-determined information on cloud-types, and even systems using emerging diffusion modeling.

Goals:

• To identify, innovate, and benchmark efficient and accurate methodologies for 3-hour nowcasting of 3D cloud and visibility fields.
• To implement 3D cloud and visibility nowcasting products in near real-time.
• To use optical flow and machine learning to nowcast cloud imagery in a computationally efficient manner (see figure below).

Figure: 3 hour infrared imagery forecast using optical flow