Cloudnet categorize and classification products from expedition ARTofMELT, Arctic Ocean, 2023

Heather Guy, Ian Brooks, Sonja Murto, Michael Tjernström, Michail Karalis, John Prytherch

Cloud and atmospheric state observations on a common grid derived from the RPG-Radiometer Physics RPG-FMCW-94, RPG-HATPRO, Vaisala ceilometer, and radiosoundings, from the Swedish icebreaker Oden during the expedition ARTofMELT (Atmospheric Rivers and the Onset of Arctic Melt) in the Fram Strait pack ice during the spring transition, 9 May to 12 June of 2023.

Data are quality controlled and processed using the cloudnetpy processing package, which includes categorization of targets, error variables, and data quality flags. The dataset also includes daily quicklooks of the target classification.

Citation

Heather Guy, Ian Brooks, Sonja Murto, Michael Tjernström, Michail Karalis, John Prytherch (2025) Cloudnet categorize and classification products from expedition ARTofMELT, Arctic Ocean, 2023. Dataset version 1. Bolin Centre Database. https://doi.org/10.17043/oden-artofmelt-2023-cloudnet-1

References

Achtert P, O’Connor EJ, Brooks IM, Sotiropoulou G, Shupe MD, Pospichal B, Brooks BJ, Tjernström M (2020) Properties of Arctic liquid and mixed-phase clouds from shipborne Cloudnet observations during ACSE 2014. Atmos. Chem. Phys. 20:14983–15002. https://doi.org/10.5194/acp-20-14983-2020

Illingworth AJ, Hogan RJ, O’Connor EJ, Bouniol D, Brooks ME, Delanoé J, Donovan DP, Eastment JD, Gaussiat N, Goddard JWF, Haeffelin M, Baltink HK, Krasnov OA, Pelon J, Piriou J-M, Protat A, Russchenberg HWJ, Seifert A, Tompkins AM, van Zadelhoff G-J, Vinit F, Willén U, Wilson DR, Wrench CL (2007) Cloudnet. Bulletin of the American Meteorological Society 88:883–898. https://doi.org/10.1175/BAMS-88-6-883

Schimmel W, Kalesse-Los H, Maahn M, Vogl T, Foth A, Garfias PS, Seifert P (2022) Identifying cloud droplets beyond lidar attenuation from vertically pointing cloud radar observations using artificial neural networks. Atmos. Meas. Tech. 15:5343–5366. https://doi.org/10.5194/amt-15-5343-2022

Tukiainen S, O’Connor E, Korpinen A (2020) CloudnetPy: A Python package for processing cloud remote sensing data. Journal of Open Source Software 5:2123. https://doi.org/10.21105/joss.02123

Guy H, Brooks I (2025) Voodoonet model used in Cloudnet categorize and classification products from expedition ARTofMELT, Arctic Ocean, 2023. Dataset version 1. Bolin Centre Database. https://doi.org/10.17043/oden-artofmelt-2023-cloudnet-voodoonet-model-1

Guy H, Brooks I, Murto S, Tjernström M, Karalis M, Prytherch J (2024) Cloud radar data from expedition ARTofMELT, Arctic Ocean, 2023. Dataset version 1. Bolin Centre Database. https://doi.org/10.17043/oden-artofmelt-2023-cloud-radar-1

Guy H, Murto S, Tjernström M, Karalis M, Prytherch J, Brooks I (2025) Cloud base heights and atmospheric backscatter observations from expedition ARTofMELT, Arctic Ocean, 2023 — cloudnetpy input. Dataset version 1. Bolin Centre Database. https://doi.org/10.17043/oden-artofmelt-2023-ceilometer-cloudnetpy-1

Murto S, Tjernström M, Brooks I, Guy H, Karalis M, Vihma T, Urbancic G, Prytherch J (2024a) Radiosonde profiles from expedition ARTofMELT, Arctic Ocean, 2023. Dataset version 1. Bolin Centre Database. https://doi.org/10.17043/oden-artofmelt-2023-radiosonde-1

Murto S, Tjernström M, Karalis M, Prytherch J (2024b) Visibility, precipitation and present weather observations from expedition ARTofMELT, Arctic Ocean, 2023. Dataset version 1. Bolin Centre Database. https://doi.org/10.17043/oden-artofmelt-2023-present-weather-1

Tjernström M, Murto S, Karalis M, Prytherch J (2024) Temperature and humidity profiles from microwave radiometer during expedition ARTofMELT, Arctic Ocean, 2023 — raw data. Dataset version 1. Bolin Centre Database. https://doi.org/10.17043/oden-artofmelt-2023-microwave-profiles-raw-1

Data description

Three types of data are provided, each consisting of 35 daily files (one per day, 9th May to 12th June 2023 inclusive):

• categorize Cloudnet level 1c categorize product in daily NetCDF (nc) files for the entire ARTofMELT campaign. (992.5 MB)
• classification Cloudnet level 1c classification product in daily NetCDF (nc) files for the entire ARTofMELT campaign. (6.6 MB)
• quicklooks Quick-look summary plots (png). (21.5 MB)

categorize

This directory contains all of the quality controlled remote sensing data and derived variables on a common grid. This includes lidar attenuated backscatter from the ceilometer, radar moments with a correction for attenuation, liquid water path from the microwave radiometer, interpolated temperature and humidity from radiosonde profiles, and horizontal visibility from the present weather sensor. Derived quantities include a probability of liquid droplet and insect presence, rainfall rate, and a bitmap of category flags that are used to determine a target categorization from the combined observations.

Variables

• All categorize variables
  • Z Radar reflectivity factor, dBZ
  • v Doppler velocity, m s-1
  • width Spectral width, m s-1
  • sldr Slanted linear depolarisation ratio, dB
  • rainfall_rate Rainfall rate, m s-1
  • latitude Latitude of site, degree_north
  • longitude Longitude of site, degree_east
  • v_sigma Standard deviation of mean Doppler velocity, m s-1
  • nyquist_velocity Nyquist velocity, m s-1
  • Z_error Error in radar reflectivity factor, dB
  • Z_sensitivity Minimum detectable radar reflectivity, dBZ
  • Z_bias Bias in radar reflectivity factor, dB
  • altitude Altitude of site, m
  • time Time UTC
  • height Height above mean sea level, m
  • radar_frequency Radar transmit frequency, GHz
  • radar_gas_atten Two-way radar attenuation due to atmospheric gases, dB
  • radar_liquid_atten Two-way radar attenuation due to liquid water, dB
  • radar_rain_atten Two-way radar attenuation due to rain, dB
  • radar_melting_atten Two-way radar attenuation due to melting ice, dB
  • quality_bits Data quality bits
  • insect_prob Insect probability
  • rain_detected Rain detected
  • category_bits Target categorization bits
  • beta Attenuated backscatter coefficient, sr-1 m-1
  • lidar_wavelength Laser wavelength, nm
  • beta_error Error in attenuated backscatter coefficient, dB
  • beta_bias Bias in attenuated backscatter coefficient, dB
  • model_time Model time UTC
  • model_height Height of interpolated radiosonde variables above mean sea level, m
  • Tw Wet-bulb temperature, K
  • temperature Temperature, K
  • pressure Pressure, Pa
  • q Specific humidity
  • uwind Zonal wind, m s-1
  • vwind Meridional wind, m s-1
  • lwp Liquid water path, kg m-2
  • lwp_error Error in liquid water path, kg m-2
  • horizontal_visibility Visibility, 10-minute average, m
  • day_of_year time as decimal day of year
  • liquid_prob Liquid probability

classification

These files contain detected cloud base and cloud top height, and the final target categorization and detection status derived from the categorization file bitmaps.

Variables

• All classification variables
  • target_classification Target classification
  • detection_status Radar and lidar detection status
  • cloud_base_height_amsl Height of cloud base above mean sea level, m
  • cloud_top_height_amsl Height of cloud top above mean sea level, m
  • cloud_base_height_agl Height of cloud base above ground level, m
  • cloud_top_height_agl Height of cloud top above ground level, m
  • altitude Altitude of site, m
  • latitude Latitude of site, degree_north
  • longitude Longitude of site, degree_east
  • time Time UTC
  • height Height above mean sea level, m
  • day_of_year time as decimal day of year

quicklooks

This directory contains summary plots (.png format) of the target categorisation and detection status from the classification files, overlaid with temperature contours from the interpolated radiosonde profiles.

Comments

Algorithm details

These Cloudnet level 1c products were generated using cloudnetpy version 1.66.10, with the following modifications:

1. The ceilometer noise correction was not implemented within the cloudnetpy processing chain but was computed separately due to non-typical format of the ceilometer data. See Guy et al. (2025) for more details.

2. Temperature threshold for insects has been increased relative to the default algorithm to 274 K.

3. A horizontal visibility field has been added based on 10-minute average data from the Murto et al. (2024b) insect classifications have been removed where visibility is < 1000 m per Achtert et al. (2020).

4. Interpolated 6-hourly radiosonde data was used for the temperature and humidity profile rather than model data.

5. The Voodoo algorithm (version 0.1.10) has been used to determine the liquid probability within each pixel. The model used within Voodoo (RPG_merged_cloudy_resampledv_trained-model.pt, Guy & Brooks (2025)) was trained on 596 days of data from 5 cloudnet sites (Leipzig, Burcharest, Neumayer, Norunda, and Hyytiala). The default Voodoo algorithm was modified so that each doppler spectra were resampled onto a common vertical velocity grid prior to training and classification with Voodoo to make the classification more robust to changes in radar chirp settings. Contact dataset author for further information and performance statistics.

Quality control

Quality control was applied to the input files before cloudnet processing, see the input data files listed below for further information.

Additional quality control was applied as part of the cloudnetpy processing chain, including corrections for radar attenuation and flagging of poor data (rain on the microwave radiometer for example), see the cloudnetpy algorithm for further details.

Usage and interpretation notes

The Cloudnet classification product is an excellent tool for providing a quick view of the state of the atmosphere and cloud profile on a given day. By combining information from multiple synergist remote sensing instruments, the cloudnet classification product provides a more complete picture of atmospheric state than any of the individual instruments alone. However the retrieval is limited by the inherent limitations and uncertainties associated with the input data, as well as various assumptions and thresholds within the cloudnet algorithm.

If you plan to use this dataset for more detailed analysis, you should also plan to fully examine the components in the categorisation file that have contributed to the ‘best-guess’ pixel classification in each case. Contacting the dataset author in advance with details of your application is strongly recommended.

In particular, note the following points:

• There is no universally applicable approach to distinguishing between cloud ice and frozen precipitation, and no attempt to do so has been implemented in this dataset. The result of this is that both cloud ice and frozen precipitation are simply labelled ‘ice’ in this dataset, and therefore multi-layer clouds may not be correctly represented.

• The ‘aerosol’ category is based on the detection of an above-noise signal in the ceilometer backscatter data that is not attributed to ice or droplets. There is no information about possible aerosol layers above the attenuation of the lidar backscatter.

• Liquid layers above lidar attenuation are identified from the shape of the radar doppler spectra distribution using the Voodoo machine learning algorithm. This methodology relies on the liquid droplet signal being identifiable in the radar doppler spectra. Large uncertainties can arise when the doppler spectra are dominated by turbulent processes or/and when the ice crystal population is very small and has a similar fall velocity to liquid droplet population. Further information about Voodoo performance on the ARTofMELT dataset is available upon request.

Associated raw data and code

1. Cloudnetpy version 1.66.10 was obtained from GitHub repository actris-cloudnet/cloudnetpy.
2. Voodoonet version 0.1.10 was obtained from GitHub repository actris-cloudnet/voodoonet.

Algorithm input data

The following datasets are used from expedition ARTofMELT, Arctic Ocean, 2023.

1. Cloud base heights and atmospheric backscatter observations — cloudnetpy input (Guy et al. 2025)
2. Cloud radar data (Guy et al. 2024)
3. The trained model used as part of the Voodoo algorithm (Guy & Brooks 2025)
4. Radiosonde profiles (Murto et al. 2024a)
5. Temperature and humidity profiles from microwave radiometer — raw data (Tjernström et al. 2024)
6. Visibility, precipitation and present weather observations (Murto et al. 2024b)

The expedition

Data were collected during the ARTofMELT 2023 expedition on board the Swedish icebreaker Oden, which was organized by the Swedish Polar Research Secretariat from 8 May through 14 June, 2023.

Data creators' roles

• Heather Guy, University of Leeds (operator and dataset creator)
• Ian Brooks, University of Leeds (operator and principal investigator)
• Sonja Murto, Stockholm University (operator)
• Michail Karalis, Stockholm University (operator)
• Michael Tjernström, Stockholm University (chief scientist; operator)
• John Prytherch, Uppsala University, is responsible for the instrument system.