Tuesday, 14 January 2025: 2:15 PM
348 (New Orleans Ernest N. Morial Convention Center)
Clouds are vital for Earth's radiative balance and affect industrial sectors like solar energy, airports, and ground-to-space optical communication. Cloud cover day and night measurement uncertainties persist due to subjective observations. Various ground instruments like radiometers and ceilometers offer improved accuracy, while infrared all-sky cameras like the Sky InSight™ offer more granular cloud cover descriptions. We aim to compare cloud cover data from these methods and suggest ways to address any disparities.
We analyze cloud cover time series from the NOAA Surfrad Radiation Budget Network (SURFRAD) Table Mountain site in Boulder, Colorado, USA, for 2023. Data are collected using a pyrgeometer (Eppley’s PIR), pyranometer (Eppley 8-48), a ceilometer (Vaisala’s CL51), and an infrared all-sky camera (Reuniwatt’s Sky InSight™). Different methodologies are applied depending on the instrument used. For the radiometers, we utilize the existing NOAA RadFlux product (Riihimaki et al. 2019). For the ceilometer, we calculate a weighted average of low cloud occurrences over the previous 30 minutes following the approaches in Wagner et al. (2015). With the Sky InSight™, binary cloud mask images are obtained by removing the clear-sky radiance from the raw image, and applying a threshold. We employ the provided cloud masks and compute the ratio of cloudy pixels over the total number of pixels with a zenith angle below 70°. Pixels are weighted by their solid angle. We also compute two other cloud fraction values using the raw binary cloud masks: a ceilometer emulation is computed using only pixels within 30° from zenith and average over 30 minutes, while a pyrgeometer emulation is computed by weighing pixels by their ground projected solid angles and averaged over 10 minutes.
The predominant readings are clear skies (0 and 1 octa) or overcast (7 or 8 octas), accounting for 70.1% of cases for the pyrgeometer, 87.8% for the ceilometer, and approximately 76.1% for the Sky InSight™ hemispherical and ground cloud covers. Camera readings are less likely to produce results of 0 or 8 octas compared to all other methods. Indeed, as per WMO recommendations, these readings require perfectly clear or overcast skies, and the camera’s wider field of view makes it more likely for it to correctly detect small clouds at high zenith angles or gaps in the cloud cover, producing readings of 1 octa instead. This interpretation is reinforced by the readings from ceilometer emulation by the camera, whose distribution more closely matches the ceilometer readings.
Overall, cross-instrument accuracy within one octa across all methods ranges between 75% and 85%, although this must be considered in the context of the high occurrence of clear and overcast skies at the site. For partial cloud cover cases (octas 2-6), cross-instrument accuracies drop to 25%-46%, with the ceilometer showing the most divergence and the camera and pyrgeometer tending to agree more often. The camera emulation of a ceilometer significantly improves agreement between the two instruments, suggesting between 25% and 35% of the ceilometer’s variance is due to the reduced field of view and averaging window.
Analysis reveals that ceilometer measurements often miss cloud edges or gaps in the cloud cover due to their narrow spatial constraints, making partial cloud cover readings less frequent. They are also more sensitive to optically thin clouds than the other instruments. Conversely, while both the camera and pyrgeometer are sensitive to the entire sky dome, the latter tends to underestimate cloud cover for thin overcast skies (altostratus), whereas the former may miss some high-altitude cirrus clouds.
We analyze cloud cover time series from the NOAA Surfrad Radiation Budget Network (SURFRAD) Table Mountain site in Boulder, Colorado, USA, for 2023. Data are collected using a pyrgeometer (Eppley’s PIR), pyranometer (Eppley 8-48), a ceilometer (Vaisala’s CL51), and an infrared all-sky camera (Reuniwatt’s Sky InSight™). Different methodologies are applied depending on the instrument used. For the radiometers, we utilize the existing NOAA RadFlux product (Riihimaki et al. 2019). For the ceilometer, we calculate a weighted average of low cloud occurrences over the previous 30 minutes following the approaches in Wagner et al. (2015). With the Sky InSight™, binary cloud mask images are obtained by removing the clear-sky radiance from the raw image, and applying a threshold. We employ the provided cloud masks and compute the ratio of cloudy pixels over the total number of pixels with a zenith angle below 70°. Pixels are weighted by their solid angle. We also compute two other cloud fraction values using the raw binary cloud masks: a ceilometer emulation is computed using only pixels within 30° from zenith and average over 30 minutes, while a pyrgeometer emulation is computed by weighing pixels by their ground projected solid angles and averaged over 10 minutes.
The predominant readings are clear skies (0 and 1 octa) or overcast (7 or 8 octas), accounting for 70.1% of cases for the pyrgeometer, 87.8% for the ceilometer, and approximately 76.1% for the Sky InSight™ hemispherical and ground cloud covers. Camera readings are less likely to produce results of 0 or 8 octas compared to all other methods. Indeed, as per WMO recommendations, these readings require perfectly clear or overcast skies, and the camera’s wider field of view makes it more likely for it to correctly detect small clouds at high zenith angles or gaps in the cloud cover, producing readings of 1 octa instead. This interpretation is reinforced by the readings from ceilometer emulation by the camera, whose distribution more closely matches the ceilometer readings.
Overall, cross-instrument accuracy within one octa across all methods ranges between 75% and 85%, although this must be considered in the context of the high occurrence of clear and overcast skies at the site. For partial cloud cover cases (octas 2-6), cross-instrument accuracies drop to 25%-46%, with the ceilometer showing the most divergence and the camera and pyrgeometer tending to agree more often. The camera emulation of a ceilometer significantly improves agreement between the two instruments, suggesting between 25% and 35% of the ceilometer’s variance is due to the reduced field of view and averaging window.
Analysis reveals that ceilometer measurements often miss cloud edges or gaps in the cloud cover due to their narrow spatial constraints, making partial cloud cover readings less frequent. They are also more sensitive to optically thin clouds than the other instruments. Conversely, while both the camera and pyrgeometer are sensitive to the entire sky dome, the latter tends to underestimate cloud cover for thin overcast skies (altostratus), whereas the former may miss some high-altitude cirrus clouds.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner