This work aims at identifying the best performing mid- and far- infrared (MIR and FIR) joint spectral interval to identify and classify clouds in the Antarctic region by mean of a machine learning algorithm. About 1700 high spectral resolution radiances, collected during 2013 by the ground based Radiation Explorer in the Far InfraRed-Prototype for Applications and Development, REFIR-PAD (Palchetti et al., 2015) at Dome C, Antarctic Plateau, are co-located with backscatter and depolarization profiles derived from a tropospheric lidar system (Ricaud et al., 2017) to pre-classify clear sky, ice clouds, or mixed phase clouds. A machine learning cloud identification and classification algorithm named CIC (Maestri et al., 2019), trained with a pre-selected set of REFIR-PAD spectra, is applied to the selected radiance dataset by assuming that no other information than the spectrum itself is known. The CIC algorithm is applied by considering different spectral intervals, in order to maximize the classification results for each class (clear sky, ice clouds, mixed phase clouds). A CIC "threat score" is defined as the classification true positives divided by the sum of true positives, false positives, and false negatives. The maximization of the threat score is used to assess the algorithm performances that span from 58% to 96% in accordance with the selected interval. The best performing spectral range is the 380-1000 cm -1 . The result, besides suggesting the importance of a proper algorithm calibration in accordance with the used sensor, highlights the fundamental role of the far infrared part of the spectrum. The calibrated CIC algorithm is then applied to a larger REFIR-PAD dataset of about 90000 spectra collected from 2012 to 2015. Some results of the full dataset cloud classification are also presented. The present work contributes to the preparatory studies for the Far-infrared Outgoing Radiation Understanding and Monitoring (FORUM) mission that has recently been selected as ESA's 9 th Earth Explorer mission, scheduled for launch in 2026. References: Maestri, T., Cossich, W., and Sbrolli, I., 2019: Cloud identification and classification from high spectral resolution data in the far infrared and mid-infrared, Atmos. Meas. Tech., 12, pp. 3521 - 3540 Palchetti, L., Bianchini, G., Di Natale, G., and Del Guasta, M., 2015: Far infrared radiative properties of water vapor and clouds in Antarctica. Bull. Amer. Meteor. Soc., 96, 1505-1518, doi: http://dx.doi.org/10.1175/BAMS- D-13-00286.1. Ricaud, P., Bazile, E., del Guasta, M., Lanconelli, C., Grigioni, P., and Mahjoub, A., 2017: Genesis of diamond dust, ice fog and thick cloud episodes observed and modelled above Dome C, Antarctica, Atmos. Chem. Phys., 17, 5221-5237, https://doi.org/10.5194/acp-17-5221-2017.
Antarctic cloud detection and classification from far and mid infrared downwelling radiance spectra: performances optimization and results
Gianluca di Natale;Luca Palchetti;Massimo Del Guasta
2020
Abstract
This work aims at identifying the best performing mid- and far- infrared (MIR and FIR) joint spectral interval to identify and classify clouds in the Antarctic region by mean of a machine learning algorithm. About 1700 high spectral resolution radiances, collected during 2013 by the ground based Radiation Explorer in the Far InfraRed-Prototype for Applications and Development, REFIR-PAD (Palchetti et al., 2015) at Dome C, Antarctic Plateau, are co-located with backscatter and depolarization profiles derived from a tropospheric lidar system (Ricaud et al., 2017) to pre-classify clear sky, ice clouds, or mixed phase clouds. A machine learning cloud identification and classification algorithm named CIC (Maestri et al., 2019), trained with a pre-selected set of REFIR-PAD spectra, is applied to the selected radiance dataset by assuming that no other information than the spectrum itself is known. The CIC algorithm is applied by considering different spectral intervals, in order to maximize the classification results for each class (clear sky, ice clouds, mixed phase clouds). A CIC "threat score" is defined as the classification true positives divided by the sum of true positives, false positives, and false negatives. The maximization of the threat score is used to assess the algorithm performances that span from 58% to 96% in accordance with the selected interval. The best performing spectral range is the 380-1000 cm -1 . The result, besides suggesting the importance of a proper algorithm calibration in accordance with the used sensor, highlights the fundamental role of the far infrared part of the spectrum. The calibrated CIC algorithm is then applied to a larger REFIR-PAD dataset of about 90000 spectra collected from 2012 to 2015. Some results of the full dataset cloud classification are also presented. The present work contributes to the preparatory studies for the Far-infrared Outgoing Radiation Understanding and Monitoring (FORUM) mission that has recently been selected as ESA's 9 th Earth Explorer mission, scheduled for launch in 2026. References: Maestri, T., Cossich, W., and Sbrolli, I., 2019: Cloud identification and classification from high spectral resolution data in the far infrared and mid-infrared, Atmos. Meas. Tech., 12, pp. 3521 - 3540 Palchetti, L., Bianchini, G., Di Natale, G., and Del Guasta, M., 2015: Far infrared radiative properties of water vapor and clouds in Antarctica. Bull. Amer. Meteor. Soc., 96, 1505-1518, doi: http://dx.doi.org/10.1175/BAMS- D-13-00286.1. Ricaud, P., Bazile, E., del Guasta, M., Lanconelli, C., Grigioni, P., and Mahjoub, A., 2017: Genesis of diamond dust, ice fog and thick cloud episodes observed and modelled above Dome C, Antarctica, Atmos. Chem. Phys., 17, 5221-5237, https://doi.org/10.5194/acp-17-5221-2017.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
