N23A-T025   TITLE: Cloud Nowcasting Data/Model Fusion

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Artificial Intelligence (AI)/Machine Learning (ML); Autonomy; General Warfighting Requirements (GWR)

OBJECTIVE: Develop an algorithmic tool for the seamless assessment of previous, current and short term forecast (0-24 h) atmospheric cloud characteristics such as ceiling, thickness, and optical depth for single and multiple overlapping cloud fields using machine learning methods to combine satellite based environmental monitoring (SBEM) analysis information, ground based remote sensing, and numerical weather prediction model fields.

DESCRIPTION: Despite increasing complexity and accuracy of weather forecast models, tools for their use in diagnosing cloud and visibility characteristics are relatively unsophisticated. Many use cases assume a relatively static cloud field based on available satellite analysis data, climatological occurrence of cloud locations, or derived numerical weather prediction output (such as relative humidity fields). Using state of the art technology, this STTR topic seeks to bridge the gap between cloud analysis by satellite and forecasts from numerical modeling by using machine learning techniques and other data fusion capabilities to improve forecast uncertainty and cloud representation through a 24 hour forecast lead-time. Effort should focus on improved post processing of cloud characteristics in numerical modeling as well as techniques to seamlessly morph a true cloud analysis field from satellite into the predicted field. Implementation should take advantage of modern software strategies, including concise data presentation and intuitive user interface to display custom visualization of clouds from any 3 dimensional angle.

PHASE I: Determine and demonstrate the technical capability to smoothly transition observed satellite observations of cloud cover into simulated cloud (or meteorological variables interpreted as cloud) forecasts. Work should identify methodological details to preserve the physical structure of observed cloud visibility characteristics, aesthetics and ease of interpretation, and incorporation of probabilistic information for metrics of uncertainty and/or alternate future scenarios. Skill with respect to aviation needs should be enumerated (e.g., accuracy of cloud ceilings, error in horizontal and vertical positioning, treatment of overlapping cloud layers) given validating data from ground lidar or other full atmosphere representation. Develop a final summary report, including literature review and overall conclusions/recommendations, to be presented at the end of this Phase. Develop a Phase II plan.

PHASE II: Expanded technical development and validation of a robust prototype system of seamless historical, current, and future cloud state. Effort should be focused on 1) maturing complex machine learning algorithms that improve forecast representation of clouds given observed satellite cloud state (including advective schemes such as optical flow, clustering of cloud types, and texture analyses), 2) adding additional data sources and expanded spatial domain to prove application to any global location, and 3) iterating on accuracy and ease of user interface in display of cloud fields. Given the need to add many different satellite platforms and forecast model data, substantial subtasks on developing generic data reader capabilities and automated metadata generation and labeling are anticipated. While the code itself may need nominal high performance computing to run, output of prototype algorithm should be capable of being visualized on a standard laptop or cellphone with modest data bandwidth (such as by tiling). It is anticipated that the prototype data output and software package will be compatible with running from open source python data analysis libraries at the conclusion of Phase II efforts. Delivery of a prototype software package and final verification report is expected at the end of this Phase.

PHASE III DUAL USE APPLICATIONS: This development will result in valuable knowledge and technology advances for the entire meteorological analysis and forecasting community, as well as downstream applications. Further follow-on Phase II efforts include expansion of observational and remote sensing datasets used to generate and validate the algorithmic tool, software refinements and hardening based on real-world operational constraints (such as data latency and drop outs, quality issues, etc.), and further tests of blending code to similar meteorological variables (such as atmospheric water constituents). DoD, civil, and private aviation will particularly benefit by having a state-of-the-art product aimed at understanding cloud visibility evolution at all stages of flight operations, from take-off through ferry and landing. Naval applications will particularly benefit by the significant increase in specific environmental data and available at any point where the Naval aircraft operations can occur. Other civil and commercial applications will benefit from enhanced data streams and software implementations for broad aviation and visibility applications, improved predictability in weather forecasts, and increased cross-over between civil and commercial satellite remote sensing activities. This effort has the potential to fill a data gap in all aspects of meteorological analysis as well as provide a foundation for additional data fusion opportunities.

REFERENCES:

1.       Mecikalski, John R., et al. "Probabilistic 0�1-h convective initiation nowcasts that combine geostationary satellite observations and numerical weather prediction model data." Journal of Applied Meteorology and Climatology 54.5 (2015): 1039-1059.

2.       Veillette, Mark S., et al. "Creating synthetic radar imagery using convolutional neural networks." Journal of Atmospheric and Oceanic Technology 35.12 (2018): 2323-2338.

3.       Wang, Chenxi, et al. "A machine-learning-based cloud detection and thermodynamic-phase classification algorithm using passive spectral observations." Atmospheric Measurement Techniques 13.5 (2020): 2257-2277.

4.       Nachamkin, Jason E., et al. "Classification and Evaluation of Stable and Unstable Cloud Forecasts." Monthly Weather Review 150.1 (2021): 81-98.

 

KEYWORDS: meteorology; weather; clouds; visibility; nowcasting; forecasting; numerical weather prediction; Satellite Based Environmental Monitoring; SBEM; satellite; remote sensing; data fusion; machine learning; optical flow; visualization; metadata; aviation

---