Understanding cirrus clouds using explainable machine learning
ETH Zurich（苏黎世联邦理工学院）；Universitat de València（瓦伦西亚大学）
Kai Jeggle, David Neubauer, Gustau Camps-Valls, Ulrike Lohmann
备注：
Presented at Climate Informatics 2023 in Cambridge; Submitted to Environmental Data Science Journal Updates Version: New version of dataset is linked. Please use that version: https://zenodo.org/record/7965381
Cirrus clouds are key modulators of Earth's climate. Their dependencies on meteorological and aerosol conditions are among the largest uncertainties in global climate models. This work uses three years of satellite and reanalysis data to study the link between cirrus drivers and cloud properties. We use a gradient-boosted machine learning model and a Long Short-Term Memory (LSTM) network with an attention layer to predict the ice water content and ice crystal number concentration. The models show that meteorological and aerosol conditions can predict cirrus properties with $R^2 = 0.49$. Feature attributions are calculated with SHapley Additive exPlanations (SHAP) to quantify the link between meteorological and aerosol conditions and cirrus properties. For instance, the minimum concentration of supermicron-sized dust particles required to cause a decrease in ice crystal number concentration predictions is $2 \times 10^{-4}$ mg m\textsuperscript{-3}. The last 15 hours before the observation predict all cirrus properties.
This study present a comparison of lidar observations of midlatitude cirrus clouds over the SIRTA observatory between 2002 and 2006 with multiple tropopauses (MT) retrieved from radiosounding temperature profiles. The temporal variability of MT properties (frequency, thickness) are discussed. Results show a marked annual cycle, with MT frequency reaching its lowest point in May (~18% occurrence of MT) and slowly rising to more than 40% in DJF. The average thickness of the MT also follows an annual cycle, going from less than 1 km in spring to 1.5 km in late autumn. Comparison with lidar observ…
The purpose of this study is to provide more insight into the role of anthropogenic cirrus formation through air traffic, by investigating the high-density European airspace over a period spanning several recent years including the start of the COVID-19 pandemic (2015-2020). Several data resources are combined, exploiting the strengths of each product within an all-encompassing framework on a high spatio-temporal resolution. Data from METEOSAT SEVIRI have been combined and validated with CALIPSO's CALIOP data to deduce temporal cirrus cloud cover variability over a rectangular region bound by …
This paper presents a simple analytical framework for the dynamic response of cirrus to a local radiative flux convergence, expressible in terms of three independent modes of cloud evolution. Horizontally narrow and tenuous clouds within a stable environment adjust to radiative heating by ascending gradually across isentropes while spreading sufficiently fast so as to keep isentropic surfaces nearly flat. More optically dense clouds experience very concentrated heating, and if they are also very broad, they develop a convecting mixed layer. Along isentropic spreading still occurs, but in the f…
Recent observations have indicated significant modulation of the Madden Julian Oscillation (MJO) by the phase of the stratospheric Quasi-Biennial Oscillation (QBO) during boreal winter. Composites of the MJO show that upper tropospheric ice cloud fraction and water vapor anomalies are generally collocated, and that an eastward tilt with height in cloud fraction exists. Through radiative transfer calculations, it is shown that ice clouds have a stronger tropospheric radiative forcing than do water vapor anomalies, highlighting the importance of incorporating upper tropospheric/lower stratospher…
University of Southern California（南加州大学）；Columbia University（哥伦比亚大学）；Tencent America（腾讯美国）
An explainable machine learning method for point cloud classification, called the PointHop method, is proposed in this work. The PointHop method consists of two stages: 1) local-to-global attribute building through iterative one-hop information exchange, and 2) classification and ensembles. In the attribute building stage, we address the problem of unordered point cloud data using a space partitioning procedure and developing a robust descriptor that characterizes the relationship between a point and its one-hop neighbor in a PointHop unit. When we put multiple PointHop units in cascade, the a…
Explainable LiDAR 3D Point Cloud Segmentation and Clustering for Detecting Airplane-Generated Wind Turbulence
Karlsruhe Institute of Technology（卡尔斯鲁厄理工学院）；ScaDS.AI, TU Dresden（德累斯顿工业大学ScaDS.AI）；Institute of Atmospheric Physics, German Aerospace Center（德国航空航天中心大气物理研究所）；RWTH Aachen University（亚琛工业大学）
The state of the atmosphere is governed by the classical laws of fluid motion and exhibits correlations in various spatial and temporal scales. These correlations are crucial to understand the short and long term trends in climate. Cirrus clouds are important ingredients of the atmospheric boundary layer. To improve future parameterization of cirrus clouds in climate models, it is important to understand the cloud properties and how they change within the cloud. We study correlations in the fluctuations of radar signals obtained at isodepths of winter and fall cirrus clouds. In particular we f…
IceCloudNet: Cirrus and mixed-phase cloud prediction from SEVIRI input learned from sparse supervision
ETH Zurich（苏黎世联邦理工学院）；European Space Agency (ESA)（欧洲空间局）；Consiglio Nazionale delle Ricerche - Istituto di Scienze Marine (CNR-ISMAR)（意大利国家研究委员会-海洋科学研究所）
Spatial buoyancy-helical distributed chaos (turbulence) in the cumulus, stratiform, stratocumulus, cirrus and cirrus mammatus clouds have been studied using results of direct numerical simulations and measurements in cloudy atmosphere. It is shown that in the considered cases the moments of helicity distribution dominate the kinetic energy spectra both in the middle of the clouds and in the cloud-top regions.
Vertically pointing Doppler radar has been used to study the evolution of ice particles as they sediment through a cirrus cloud. The measured Doppler fall speeds, together with radar-derived estimates for the altitude of cloud top, are used to estimate a characteristic fall time tc for the `average' ice particle. The change in radar reflectivity Z is studied as a function of tc, and is found to increase exponentially with fall time. We use the idea of dynamically scaling particle size distributions to show that this behaviour implies exponential growth of the average particle size, and argue t…
The stable isotopologues of water have been used in atmospheric and climatic studies for over fifty years, because the temperature-dependent preferential condensation of heavy isotopologues during phase changes makes them useful diagnostics of the hydrological cycle. However, the degree of preferential condensation has never been directly measured at temperatures below 233 K (-40$^{\circ}$C), conditions of cirrus formation in the atmosphere and routinely observed at surface elevation in polar regions. (Models generally assume an extrapolation from the warmer experiments of Merlivat and Nief, 1…
We propose an approximation for the total ice nucleation rate J=J(T,R) in supercooled water droplets as a function of both droplet radius R and temperature T, taking account of both volume-based and surface-stimulated nucleation modes. Its crucial idea is that, even in the surface-stimulated mode crystal nuclei initially emerge (as sub-critical clusters) homogeneously in the sub-surface layer, not "pseudo-heterogeneously" at the surface. This mode is negligible in large droplets, but becomes increasingly important with decreasing droplet size and is dominant in small droplets. The crossover dr…
The accuracy and explainability of data-driven nowcasting models are of great importance in many socio-economic sectors reliant on weather-dependent decision making. This paper proposes a novel architecture called Small Attention Residual UNet (SAR-UNet) for precipitation and cloud cover nowcasting. Here, SmaAt-UNet is used as a core model and is further equipped with residual connections, parallel to the depthwise separable convolutions. The proposed SAR-UNet model is evaluated on two datasets, i.e., Dutch precipitation maps ranging from 2016 to 2019 and French cloud cover binary images from …
A Low Humidity and Temperature Profiling (LHATPRO) microwave radiometer, manufactured by Radiometer Physics GmbH (RPG), is used to monitor sky conditions over ESO's Paranal observatory in support of VLT science operations. In addition to measuring precipitable water vapour (PWV) the instrument also contains an IR camera measuring sky brightness temperature at 10.5 μm. Due to its extended operating range down to -100 °C it is capable of detecting very cold and very thin, even sub-visual, cirrus clouds. We present a set of instrument flux calibration values as compared with a detrended fluctuati…
