Thursday, 16 January 2025
Hall C (New Orleans Ernest N. Morial Convention Center)
Free Space Optical Communication (FSOC) provides critical advantages over radio
frequency communications, especially with increased data rates and security. However,
establishing an optical connection between space and the ground can be impeded by atmospheric
attenuation and distortion, primarily due to clouds. Therefore, a cloud free line of sight (CFLOS)
is needed to complete the downlink. On-site ground instruments, such as an infrared cloud
imager (ICI), are installed to provide decision makers with situational awareness and to inform
short-term forecasts of CFLOS. The ICI is a fully calibrated, passive infrared instrument with
nearly horizon to horizon coverage of the sky that provides sky radiance data at each pixel within
the sky dome. Using data from the ICI, a UNet Cloud Analysis (UCA) was developed and
deployed in 2021. The UCA is optimized to identify and discriminate clouds from background
radiance measurement in highly variable atmospheric conditions. Furthermore, to aid decision
makers beyond real-time cloud analyses and to predict the viability of FSOC in the short term,
the UNet Cloud Forecasting System (UCFS) was developed to forecast cloud location out to 10
minutes in the future based on 8 months of UCA data. This initial UCFS model, improved upon
persistence at every timestep and proved skillful in the initial prediction timesteps. However,
predictions in the later timesteps became less confident and lacked continuous movement of
clouds.
These results provided the motivation to improve the UCFS, particularly for forecasts
beyond 5 minutes in the future. Utilizing additional data collected since 2021, the UCFS was
retrained using 2 years of UCA data, improved data mining techniques, and more in-depth testing
and validation. Analyses showed substantial performance increases at all timesteps and
continuous movement of clouds throughout the forecast window, particularly at the 10-minute
prediction timestep, which outperformed the initial UCFS by 10% when comparing critical
success index (CSI) score. Given this increase in performance, both qualitatively and
quantitively, several optimization studies were performed to determine if the forecast length
could be extended beyond 10 minutes while maintaining performance accuracy. This effort
resulted in the reduction of the number of input UCA images from 10 to 5 and increased the
output forecast to 20 minutes. UCFS performance remained comparable to the previous versions
and showed no substantial drop-offs in CSI or accuracy beyond the 10-minute forecast. When
comparing CSI scores between the UCFS and persistence, the UCFS increased the CSI score by
12%. Qualitatively, the predictions are skillful and discriminative at all timesteps. By extending
the prediction beyond 10 minutes, decision makers can better coordinate link handovers between
FSOC sites, which can take upwards of 5 minutes, and are provided with greater situational
awareness at a time and spatial resolution traditional NWP and persistence forecasts are unable
to provide.
To conclude, this presentation will discuss and overview the capabilities and
improvements of the UCFS, the process to optimize its training, and how decision makers can
utilize the system to gain situational awareness for FSOC missions.
frequency communications, especially with increased data rates and security. However,
establishing an optical connection between space and the ground can be impeded by atmospheric
attenuation and distortion, primarily due to clouds. Therefore, a cloud free line of sight (CFLOS)
is needed to complete the downlink. On-site ground instruments, such as an infrared cloud
imager (ICI), are installed to provide decision makers with situational awareness and to inform
short-term forecasts of CFLOS. The ICI is a fully calibrated, passive infrared instrument with
nearly horizon to horizon coverage of the sky that provides sky radiance data at each pixel within
the sky dome. Using data from the ICI, a UNet Cloud Analysis (UCA) was developed and
deployed in 2021. The UCA is optimized to identify and discriminate clouds from background
radiance measurement in highly variable atmospheric conditions. Furthermore, to aid decision
makers beyond real-time cloud analyses and to predict the viability of FSOC in the short term,
the UNet Cloud Forecasting System (UCFS) was developed to forecast cloud location out to 10
minutes in the future based on 8 months of UCA data. This initial UCFS model, improved upon
persistence at every timestep and proved skillful in the initial prediction timesteps. However,
predictions in the later timesteps became less confident and lacked continuous movement of
clouds.
These results provided the motivation to improve the UCFS, particularly for forecasts
beyond 5 minutes in the future. Utilizing additional data collected since 2021, the UCFS was
retrained using 2 years of UCA data, improved data mining techniques, and more in-depth testing
and validation. Analyses showed substantial performance increases at all timesteps and
continuous movement of clouds throughout the forecast window, particularly at the 10-minute
prediction timestep, which outperformed the initial UCFS by 10% when comparing critical
success index (CSI) score. Given this increase in performance, both qualitatively and
quantitively, several optimization studies were performed to determine if the forecast length
could be extended beyond 10 minutes while maintaining performance accuracy. This effort
resulted in the reduction of the number of input UCA images from 10 to 5 and increased the
output forecast to 20 minutes. UCFS performance remained comparable to the previous versions
and showed no substantial drop-offs in CSI or accuracy beyond the 10-minute forecast. When
comparing CSI scores between the UCFS and persistence, the UCFS increased the CSI score by
12%. Qualitatively, the predictions are skillful and discriminative at all timesteps. By extending
the prediction beyond 10 minutes, decision makers can better coordinate link handovers between
FSOC sites, which can take upwards of 5 minutes, and are provided with greater situational
awareness at a time and spatial resolution traditional NWP and persistence forecasts are unable
to provide.
To conclude, this presentation will discuss and overview the capabilities and
improvements of the UCFS, the process to optimize its training, and how decision makers can
utilize the system to gain situational awareness for FSOC missions.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner