OVERCAST

Optical Variability Evaluation of Regional Cloud Asymmetries in Space and Time

The Project
Science Demonstration Application Validation Visualization Related Programs
Staff Log In
Contact

OVERCAST Applications



Multilayer clouds: The current CBH/CGT products are still optimized to single-layer clouds. Opportunities for improvement of techniques for multi-layered systems will involve satellite/model fusion and ML techniques.

We have applied ML/AI to detect the presence of low clouds using ABI bands and NWP model humidity fields trained against the “truth” from co-located CloudSat radar and CALIPSO lidar observations.

We will attempt to expand this research for a better representation of lower level cloudiness over the globe by applying ML models to various sensors through channel adjustments and input optimization.



Figure: Improved Cloud Layers adopting machine learning (ML) approaches which produces a new high+low cloud layer category (pink) and more deep cloud pixels revealing hidden lower level cloud presence under top layers. Real-time display for improved ML-based Cloud Layer products with both GOES-16 and GOES-17 ABI available in CIRA’s SLIDER.


Nighttime clouds: we will adopt a data-fusion approach, utilizing VIIRS Day/Night Band (DNB) lunar reflectance with numerical weather prediction (NWP) model, IR, and microwave for nighttime retrievals. ML methods will be explored to calculate cloud water path for improved nighttime CGT. We will also attempt to develop synthetic Day/Night Band adopting AI/ML techniques, leveraging CIRA’s Proxy VIS research.

Transition to GeoIPS: Products developed as a part of this project are required to be transitioned to the Navy through integration with the Geolocated Information Processing System (GeoIPS).

AI-based cloud water profile estimation: Multi-output learning, in particular, holds promise as a method of retrieving entire profiles of hydrometeor occurrence from passive satellite imagery, and we will explore ways to use this to supplement our current approach in problematic multilayer situations.

AI-based global hi-resolution microwave data and synthetic radar data: We have developed deep learning approaches to produce hi-resolution microwave data (89-GHz Imagery generation from geostationary satellites) and synthetic radar reflectivity (GREMLIN that fuses ABI and GLM data). We will expand these efforts to achieve global coverage.

D-CFLOS: The Probability of Cloud-Free-Line-of-Sight (PCFLOS) is a tool to calculate the probability that a visible line-of-sight exists through the atmosphere between an observer and a target potentially obstructed by clouds, which is an effective measure to characterize the cloud environment and determine the safest altitude to fly while still maintaining a view of ground targets.

Leveraging CIRA’s research to develop global PCFLOS climatologies using CloudSat/CALIPSO observations, we will introduce a new deterministic version based on the 3D cloud structure information. The “D-CFLOS” estimate, will be practically instantaneous on a global scale, and can be used to support model forecasts of cloud cover (e.g., model verification and initialization).

Short-term advection: We will explore the potential of CIRA’s Dense Optical Flow research in the context of producing enhanced temporal resolution 3D fields between the standard analysis times, as well as potential short-term advection of the analysis field utilizing model data fusion.