Optimal Sensing of Tropical Cyclones (TCs) by Constellation of Low Earth Orbiting Satellites, Guided by Numerical Weather Prediction Model Ensemble Track Forecasts and Assimilation of Observed/ Missed TC Center Position Measurements

Abstract:

Maximizing the return of scientific-data from an observing system allows for efficient use of constrained set of resources (satellites, instruments, ground-stations, etc.). When the subject of study is a fast-developing phenomenon such as tropical cyclones (TCs), the problem becomes more challenging due to the added time-constraint. TCs are localized events which are moving and are difficult to track from LEO satellites. An optimal remote sensing strategy can be carried out if the observing system can learn from the gathered observations (both success and failures) and use this knowledge to command (time and location) future observations. Due to the time-constraint this process needs to be executed onboard the satellites, which requires the algorithm to meet a tight processing computational budget.

In this work we describe such a potential onboard algorithm which can be used to track TCs. A key enabling component of the algorithm is the use of ensemble forecasts from Numerical Weather Prediction models. The set of ensemble forecasts are used as possible TC tracks and the algorithm is used to converge upon the true track of the TC (possibly outside the set of ensembles) by assimilating potentially aperiodic (possibly missing) noisy measurements of the TC center position.

A representative LEO satellite constellation with 3 planes, 8 satellites per plane was considered as the observing system in this study. The algorithm was tested and analyzed with the Global Ensemble Forecasting System data (GEFS) and National Hurricane Center data for the 2018 year hurricanes within the Atlantic basin. Compared to a baseline method which uses the GEFS issued mean ensemble track (AEMN) for forecasting and no data assimilation, the proposed algorithm exhibited positive forecast skill for more than 290 test cases (in a total of 318 test cases) over forecast periods spanning 6–48 h. The skill is seen to improve with lengthening forecast periods, with five test cases showing greater than 75% skill for a forecast period of 6 h to 247 test cases for the forecast period of 48 h. The low computational load and the use of only the TC center measurements from the payload data make this algorithm a viable option for practical implementation.