Post-simulation diagnostics of microphysical process rates with AI

Project ID: MICRO

Introduction

In this project, we will develop AI-based tools for post-simulation diagnostics of variables which are normally not stored in a weather and climate model. Many variables are calculated at interim steps during a weather and climate simulation. But because of time and space constraints, it has to be decided which of these variables to save for later analysis. Most of the time the first victims of these constraints are process rates, e.g. cloud microphysical processes like condensation and freezing.
These processes are important for understanding feedback loops like cloud adjustments to anthropogenic or natural perturbations, but fall often short due to the constraints mentioned above. We will develop methods to estimate the values of these process rates based on standard output variables and auxiliary information. For this, training and evaluation datasets will be generated with the flexible atmospheric model ICON at convection resolving grid spacing by simulations of several days with different weather situations.

Project description

Climate and weather models need to represent key state variables, and also physical mechanisms, realistically in order to produce reliable short-term forecasts and long-term projections. There is a strong need for advanced process-oriented diagnostics (Maloney et al., 2019) to evaluate their performance. Clouds and cloud feedbacks are a large factor of uncertainty in the simulation of past, present and future climate (Boucher et al., 2013).

For a true process-based analysis of cloud microphysics, aerosol-cloud interactions and pathways of precipitation formation in numerical models, information on microphysical process rates such as condensation, freezing and hydrometeor growth is indispensable. The susceptibility of clouds and precipitation to perturbations (e.g., additional aerosols through anthropogenic emissions) can only be assessed if the microphysical pathways are understood. Snapshots of state variables such as the mixing ratios of hydrometeors are not sufficient to elucidate the microphysical pathways. However, microphysical process rates are not commonly output because they would comprise ~20-50 threedimensional variables (depending on the desired level of detail), with two variables for each arrow on representing its mass and number. This number of output variables easily triples the required storage compared to the default output and therefore limits the feasibility to short case studies and relatively small numbers of sensitivity experiments (e.g., Barthlott and Hoose, 2018).

The microphysical processes are heavily parameterized. Due to the sequence of processes and interdependencies, the standard output variables are not sufficient to naively recalculate the process rates based on excerpts from the original model code. Preliminary attempts for ice formation processes show deviation of several orders of magnitude for many cases, and frequent “misses” (i.e. where the recalculation results in a process rate of zero, while the actual rate in the model has a high value). For other processes, attempts with naïve recalculation failed. The reasons for this will be explored as part of the project, but are likely due to the nonlinear and threshold-based nature of the microphysical processes in clouds.

What we propose here is to develop a method supported by artificial intelligence to estimate the microphysical process rates after simulation with the ICOsahedral Nonhydrostatic (ICON) model in a limited area configuration (Heinze et al., 2019), based on a limited set of standard output variables.

References

Boucher, O., … Hoose, C., et al, (2013): Clouds and Aerosols. In: Climate Change 2013: The Physical Science Basis (IPCC Assessment Report 5).

Carro-Calvo, L., C. Hoose, M. Stengel, and S. Salcedo-Sanz (2016): Cloud Glaciation Temperature Estimation from Passive Remote Sensing Data with Evolutionary Computing, J. Geophys. Res. Atmos., 121, 13591-13608, doi:10.1002/2016JD025552

Heinze, R., … Hoose, C., … et al. (2017): Large-eddy simulations over Germany using ICON: a comprehensive evaluation. Q.J.R. Meteorol. Soc. doi:10.1002/qj.2947

Maloney, E. D., et al. (2019). Process-Oriented Evaluation of Climate and Weather Forecasting Models, Bulletin of the American Meteorological Society, 100(9), 1665-1686. https://journals.ametsoc.org/view/journals/bams/100/9/bams-d-18-0042.1.xml

Neelin, J. D., et al. (2020). Machine Learning Based Process-oriented Earth System Model Evaluation, AGU Fall Meeting 2020, https://ui.adsabs.harvard.edu/abs/2020AGUFMGC105..01N/abstract

Reinhardt, T., and A. Seifert, 2006: A three-category ice scheme for LMK. COSMO Newsl., 6, 115–120.

Götz, M., … Streit, A. (2020): HeAT - A Distributed and GPU-accelerated Tensor Framework for Data Analytics, 2020 IEEE International Conference on Big Data (Big Data), IEEE, pp. 276-287, https://doi.org/10.1109/BigData50022.2020.9378050