Adaptive, Model-driven Observation for Earth Science

Abstract:

We describe an initial effort to apply adaptive sensing to the study of large scale storms such as hurricanes/typhoons. In this effort, we conduct simulations of the storm. We then use these simulations to drive analysis to estimate the impact of a range of sensing actions on improvement of the model forecast ability, characterizing this as an estimation of a sensing utility function. A sensing planner then develops a sensing plan using this utility information, attempting to maximize utility gain that is subject to operational constraints of the sensor platforms (marine, aerial, space, etc.). These observations are acquired and the cycle continues. The core principle of this approach is that deliberative direction of sensor assets will be more effective than current undirected or ad hoc strategies for sensing and therefore result in improved model accuracy and consequently science. Models of the hurricane can be used to estimate the sensitivity of model outputs of interest to more local measurable quantities [1]. Below we show a preliminary utility estimation based univariate sensitivity analysis of a forecast metric (minimum sea level pressure) to local sea level pressure measurements. Sensor Planning: In the prototype under implementation, the planner/resource allocator then allocates measurement, with the objective of optimizing measurements according to the provided utility function. We are currently examining sensor allocation strategies for both space assets (Geostationary with pointing capability within a fixed field of regard, or Low Earth Orbit with a field of regard related to a fixed overflight path) as well as Aerial and In-situ assets with a deployment, path planning, and possibly recovery constraints. Status: This project is in early stages of prototyping and is still developing methods to evaluate any efficiency gains from this approach. We plan to evaluate the approach using a data denial experimental setup using historical and simulation data. Acknowledgments: This work was performed at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. © California Institute of Technology 2020. All Rights Reserved. References: [1] Torn D. R. and Hakim G. J. (2008) Ensemble-Based Sensitivity Analysis in AMS.