The formation of drizzle and rain via drop collision-coalescence (“warm rain” initiation) is a key component of Earth system models. This is typically represented using microphysics schemes that only predict a few variables to describe the drop population. When developing these relatively simple “bulk” schemes, choices must be made regarding which cloud and rain variables to predict and how to calculate the micro-scale physical processes affecting drops. These structural choices need to be simple enough so that schemes can run efficiently in three-dimensional models. This talk will describe an investigation of the accuracy of bulk schemes using different sets of predicted variables and different methods, including machine learning, for representing drop collision-coalescence. The goal is to assess how accurately these schemes simulate the formation of raindrops starting from an initial population of (small) cloud droplets, relative to a detailed reference microphysical model. Using predicted variables that represent the properties of drops of all sizes — the single-category approach — leads to much greater accuracy than the traditional two-category approach with separate variables for cloud (small drops) and rain (large drops), despite both approaches utilizing the same number of predicted variables (4). A sub-optimal choice of the predicted variables, as in the two-category approach, limits overall accuracy even when the coalescence rate calculation itself is highly accurate. This places an upper limit on accuracy using machine-learned rates even when the rate calculation is perfectly emulated. On the other hand, using a neural network to obtain coalescence rates with the single-category approach gives accurate results for a reasonable computational cost. Furthermore, only 3 predicted variables are needed for relatively accurate solutions (at least as accurate as the best-performing two-category schemes with 4 predicted variables). This result is consistent with Lamb et al. (2024, JAMES), who used a variational auto-encoder to show that the number of intrinsic dimensions for the problem of warm rain initiation is roughly 3. Prospects for utilizing the single-category approach in Earth system models will be discussed, including challenges owing to the relatively coarse spatial resolution and long time step (~5 min) of these models. Finally, I will highlight recent developments of the Lagrangian “super-droplet” method for modeling microphysics, which tracks representative clouds and raindrops in the modeled flow, and how this method can be used to further improve bulk single category schemes. This includes accounting for turbulent enhancement of drop collision-coalescence, which may be critical for rain initiation in some cloud regimes.

Bio:

Hugh Morrison is a Senior Scientist in the Mesoscale and Microscale Meteorology Laboratory (MMM) at NSF NCAR. His work spans a range of topics mainly involving clouds, precipitation, and atmospheric dynamics, primarily from numerical modeling and theoretical perspectives. Research areas include cloud microphysics, cloud dynamics, aerosol-cloud interactions, and convective dynamics. He has developed or co-developed several microphysics schemes used in a variety of models, including WRF, CESM, E3SM, and GEM, among others. In his spare time, he enjoys hanging out with his 9-year-old daughter, working out, and reading about 19th and 20th-century history.

Time: 12:00 pm EST

Free!

Registration