1- 1Institute of Meteorology and Climate Research Troposphere Research (IMKTRO), Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
- *A full list of authors appears at the end of the abstract
Cloud ice forms primarily through immersion freezing in the mixed-phase regime between 0°C and -38°C. Immersion freezing is the nucleation of a liquid cloud droplet with an immersed ice nucleating particle (INP), which reduces the energy barrier for cloud ice formation. INPs are rare aerosols, and due to measurement challenges and limitations in instrument capabilities, the availability of atmospheric observations of INPs remains scarce. Thus, obtaining a global distribution of INPs using observations has so far been challenging.
Here, we will show that, through the use of the gradient boosting machine learning algorithm XGBoost, we can predict a realistic global distribution of INPs based on temperature and Copernicus Atmosphere Monitoring Service (CAMS) reanalysis aerosols. The aerosols included are three modes of dust and sea salt, sulfate, and anthropogenic black and organic carbon, both hydrophilic and hydrophobic components. We further add a land mask, a binary identifier to contrast the oceans from land. Aerosols are collocated with about 40 observed INP datasets, sparsely distributed across the globe. Approximately 85% of the data are land-based locations.
The XGBoost model performs well. Predicted regional INP spectra with temperature using CAMS four-year climatology show a good agreement with the observations, with values within one order of magnitude for most regions. Antarctica is an outlier, and a large model bias is obtained. Spatially, the XGBoost predicts a realistic pattern with peaks over deserts and lower values across the oceans. Model sensitivity to hyperparameters reveals large variations in predicted INPs over the Arabian Peninsula and North Africa, followed by Antarctica. The presented approach can act as a cost-effective immersion freezing parameterisation in global and regional weather and climate models. To this end, some preliminary results using the ICOsahedral Non-hydrostatic (ICON) model with this new parameterisation will be shown.
Alexander Böhmländer, alexander.boehmlaender@kit.edu, Institute of Meteorology and Climate Research Atmospheric Aerosol Research (IMKAAF), Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, Ottmar Möhler, ottmar.moehler@kit.edu, Institute of Meteorology and Climate Research Atmospheric Aerosol Research (IMKAAF), Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, Pia Bogert, pia.bogert@kit.edu, Institute of Meteorology and Climate Research Atmospheric Aerosol Research (IMKAAF), Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, Heike Wex, wex@tropos.de, Leibniz Institute for Tropospheric Research, Leipzig, Germany, Oliver Eckermann, eckermann@tropos.de, Leibniz Institute for Tropospheric Research, Leipzig, Germany, Christian Tatzelt, christian.tatzelt@tropos.de, Leibniz Institute for Tropospheric Research, Leipzig, Germany, Franziska Vogel, f.vogel@isac.cnr.it, Institute of Atmospheric Sciences and Climate (ISAC), National Research Council (CNR), Bologna, Italy, Mark D. Tarn, m.d.tarn@leeds.ac.uk, School of Earth and Environment, University of Leeds, Leeds, UK, Ross Herbert, R.J.Herbert@leeds.ac.uk, Institute for Climate and Atmospheric Science, University of Leeds, Leeds, UK, Ping Tian, tianping@bj.cma.gov.cn, Beijing weather modification center, Beijing , China, Xianda Gong, gongxianda@westlake.edu.cn, Research Center for Industries of the Future, School of Engineering, Westlake University, Hangzhou, China, Jie Chen, jie.chen@env.ethz.ch, Institute for Atmospheric and Climate Science, ETH Zürich, Zürich, Switzerland, Jessie M. Creamean, jessie.creamean@colostate.edu, Department of Atmospheric Science, Colorado State University, Fort Collins, CO, USA, Paul J. DeMott, Paul.Demott@colostate.edu, Department of Atmospheric Science, Colorado State University, Fort Collins, CO, USA, Naruki Hiranuma, smoon@utep.edu, University of Texas at El Paso, El Paso, Texas, USA, Christina S. McCluskey, cmcclus@ucar.edu, NSF National Center for Atmospheric Research, Boulder, CO, USA, Zoé Brasseur, zoe.brasseur@helsinki.fi, 1) Institute for Atmospheric and Earth System Research/Physics, Faculty of Science, University of Helsinki, Helsinki, Finland, 2) SIOS Knowledge Centre, Longyearbyen, Norway, André Welti, andre.welti@fmi.fi, Finnish Meteorological Institute, Helsinki, Finland, Yutaka Tobo, tobo.yutaka@nipr.ac.jp, 1) National Institute of Polar Research, Tachikawa, Tokyo, Japan, 2) Graduate Institute for Advanced Studies, SOKENDAI, Tachikawa, Tokyo, Japan, Elise K. Wilbourn, elise.wilbourn@gmail.com, 1) Department of Life, Earth, and Environmental Sciences, West Texas A&M University, Canyon, TX, USA, 2) Bioresource and Environmental Security, Sandia National Laboratories, Livermore, CA, USA, Corinna Hoose, corinna.hoose@kit.edu, Institute of Meteorology and Climate Research Troposphere Research (IMKTRO), Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
How to cite: Wallentin, G. and the Modelling INPs Team: Modelling a Global Distribution of Ice Nucleating Particles using Machine Learning, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-12119, https://doi.org/10.5194/egusphere-egu26-12119, 2026.
