Peter Spichtinger's research while affiliated with Johannes Gutenberg-Universität Mainz and other places

Kinetic process models are widely applied in science and engineering, including atmospheric, physiological and technical chemistry, reactor design, or process optimization. These models rely on numerous kinetic parameters such as reaction rate, diffusion or partitioning coefficients. Determining these properties by experiments can be challenging, especially for multiphase systems, and researchers often face the task of intuitively selecting experimental conditions to obtain insightful results. We developed a numerical compass (NC) method that integrates computational models, global optimization, ensemble methods, and machine learning to identify experimental conditions with the greatest potential to constrain model parameters. The approach is based on the quantification of model output variance in an ensemble of solutions that agree with experimental data. The utility of the NC method is demonstrated for the parameters of a multi-layer model describing the heterogeneous ozonolysis of oleic acid aerosols. We show how neural network surrogate models of the multiphase chemical reaction system can be used to accelerate the application of the NC for a comprehensive mapping and analysis of experimental conditions. The NC can also be applied for uncertainty quantification of quantitative structure–activity relationship (QSAR) models. We show that the uncertainty calculated for molecules that are used to extend training data correlates with the reduction of QSAR model error. The code is openly available as the Julia package KineticCompass. Graphical Abstract

This study investigates the influence of relative humidity with respect to ice on the extratropical tropopause inversion layer (TIL). Initially, measurements from radiosondes at a location in Germany were compared with the ERA5 reanalysis data from the ECMWF at the same geographic location. A high level of agreement was observed, with the expected limitation that ERA5 cannot resolve sharp changes in variables like humidity and stability at the tropopause as finely. When examining the TIL with respect to mean relative humidity over ice in the upper troposphere, a clear relationship with stability becomes evident. Moister profiles, on average, exhibit significantly higher maximum values of the Brunt-Väisälä frequency N 2, indicating a more stable stratification of the tropopause in these cases. This result holds true in both radiosonde measurements and ERA5 data. Considering the thickness of the TIL layer, an inverse pattern emerges. In this case, moister and more stable TILs exhibit a lower thickness. The strong agreement between radiosondes and ERA5 allows for geographical and seasonal analyses using ERA5 data alone. These analyses reveal consistent relationships in various extratropical regions of the Northern Hemisphere under different meteorological conditions. Differences in the strength of the dependence of TIL properties on relative humidity over ice are evident.

We present an asymptotic approach for the systematic investigation of the effect of gravity waves (GW) on ice clouds formed through homogeneous nucleation. In particular, we consider high- and mid-frequency GW in the tropopause region driving the formation of ice clouds, modeled with a double-moment bulk ice microphysics scheme. The asymptotic approach allows for identifying reduced equations for self-consistent description of the ice dynamics forced by GW including the effects of diffusional growth and nucleation of ice crystals. Further, corresponding analytical solutions for a monochromatic GW are derived under a single-parcel approximation. The results provide a simple expression for the nucleated number of ice crystals in a nucleation event. It is demonstrated that the asymptotic solutions capture the dynamics of the full ice model and accurately predict the nucleated ice crystal number. The present approach is extended to allow for superposition of GW, as well as, for variable ice crystal mean mass in the deposition. Implications of the results for an improved representation of GW variability in cirrus parameterizations are discussed.

Kinetic process models are widely applied in science and engineering, including atmospheric, physiological and technical chemistry, reactor design, or process optimization. These models rely on numerous kinetic parameters such as reaction rate, diffusion or partitioning coefficients. Determining these parameters by experiments can be challenging, especially for multiphase systems, and researchers often face the task of intuitively selecting experimental conditions to obtain insightful results. We developed a kinetic compass (KC) method that integrates kinetic models, global optimization, ensemble methods, and machine learning to identify experimental conditions with the greatest potential to constrain kinetic parameters. The approach is based on the quantification of model output variance in an ensemble of solutions that agree with experimental data. The utility of the KC method is demonstrated for the kinetic parameters in a multi-layer model describing the heterogeneous ozonolysis of oleic acid aerosols. We show how neural network surrogate models of the multiphase chemical reaction system can be used to accelerate the application of the kinetic compass for a comprehensive mapping and analysis of experimental conditions. The code is openly available and can be adapted to various types of process models.

We present a novel method to identify and locate weather fronts at various pressure levels to create a three dimensional structure using weather data located at the North Atlantic. It provides statistical evaluations regarding the slope and weather phenomena correlated to the identified three dimensional structure. Our approach is based on a deep neural network to locate 2D surface fronts first, which are then used as an initialization to extend them to various height levels. We show that our method is able to detect frontal locations between 500 hPa and 1000 hPa.KeywordsWeather PredictionAtmospheric PhysicsDeep LearningClimate Change

We present an asymptotic approach for the systematic investigation of the effect of gravity waves (GW) on ice clouds formed through homogeneous nucleation. In particular, we consider high- and mid-frequency GW in the tropopause region driving the formation of ice clouds, modeled with a double-moment bulk ice microphysics scheme. The asymptotic approach allows for identifying reduced equations for self-consistent description of the ice dynamics forced by GW including the effects of diffusional growth and nucleation of ice crystals. Further, corresponding analytical solutions for a monochromatic GW are derived under a single-parcel approximation. It is demonstrated that the asymptotic solutions capture to a high accuracy the dynamics of the reduced ice model during multiple nucleation events and provide a parameterization for the nucleated number of ice crystals. Extension of the present approach to allow for variable mean mass of the ice crystal distribution, as well as, implications for representation of GW variability in cirrus parameterizations are discussed.

Homogeneous freezing of solution droplets is an important pathway of ice formation in the tropopause region. The nucleation rate can be parameterized as a function of water activity, based on empirical fits and some assumptions on the underlying properties of super-cooled water, although a general theory is missing. It is not clear how nucleation events are influenced by the exact formulation of the nucleation rate or even their inherent uncertainty. In this study we investigate the formulation of the nucleation rate of homogeneous freezing of solution droplets (1) to link the formulation to the nucleation rate of pure water droplets, (2) to derive a robust and simple formulation of the nucleation rate, and (3) to determine the impact of variations in the formulation on nucleation events. The nucleation rate can be adjusted, and the formulation can be simplified to a threshold description. We use a state-of-the-art bulk ice microphysics model to investigate nucleation events as driven by constant cooling rates; the key variables are the final ice crystal number concentration and the maximum supersaturation during the event. The nucleation events are sensitive to the slope of the nucleation rate but only weakly affected by changes in its absolute value. This leads to the conclusion that details of the nucleation rate are less important for simulating ice nucleation in bulk models as long as the main feature of the nucleation rate (i.e. its slope) is represented sufficiently well. The weak sensitivity of the absolute values to the nucleation rate suggests that the amount of available solution droplets also does not crucially affect nucleation events. The use of only one distinct nucleation threshold function for analysis and model parameterization should be reinvestigated, since it corresponds to a very high nucleation rate value, which is not reached in many nucleation events with low vertical updrafts. In contrast, the maximum supersaturation and thus the nucleation thresholds reached during an ice nucleation event depend on the vertical updraft velocity or cooling rate. This feature might explain some high supersaturation values during nucleation events in cloud chambers and suggests a reformulation of ice nucleation schemes used in coarse models based on a purely temperature-dependent nucleation threshold.

While thunderstorms can pose severe risks to property and life, forecasting remains challenging, even at short lead times, as these often arise in meta-stable atmospheric conditions. In this paper, we examine the question of how well we could perform short-term (up to 180 min) forecasts using exclusively multi-spectral satellite images and past lighting events as data. We employ representation learning based on deep convolutional neural networks in an “end-to-end” fashion. Here, a crucial problem is handling the imbalance of the positive and negative classes appropriately in order to be able to obtain predictive results (which is not addressed by many previous machine-learning-based approaches). The resulting network outperforms previous methods based on physically based features and optical flow methods (similar to operational prediction models) and generalizes across different years. A closer examination of the classifier performance over time and under masking of input data indicates that the learned model actually draws most information from structures in the visible spectrum, with infrared imaging sustaining some classification performance during the night.