A document by Walter N Meier. Click on the document to view its contents.

Accurately predicting tropical cyclone (TC) activities is important for both hazard mitigation and adaptation. However, the number of TC genesis simulated by numerical models is sensitive to the model's horizontal resolution and how convection is being represented. A super-parameterized (SP) general circulation model (GCM), in which convection and sub-grid processes are explicitly simulated using a cloud-resolving microscale model embedded in each GCM grid column, but in a 2-D configuration, can achieve an explicit representation of convective processes while maintaining an acceptable computational cost. Here, we compared TC genesis using the traditionally convective parametrized Community Atmosphere Model v5.0 (using Zhang and McFarlane (1995) deep convective scheme and the University of Washington shallow convective scheme, hereafter referred to as CPCAM) with the SP CAM v5.0 (SPCAM); in particular, aquaplanet experiments with zonally symmetric sea surface temperature and perpetual summer insolation were conducted using the two models. It was found that SPCAM TC genesis was 3-4 times greater than that in CPCAM. Storm’s wind-pressure relationship was also improved, and they were stronger in SPCAM. Additionally, more “non-rotating convective clusters” and “TC seeds” were found in SPCAM compared to CPCAM. More frequent TC genesis in SPCAM was primarily related to weaker vertical wind shear (VWS) and secondarily to stronger 850-hPa absolute vorticity. Reduced VWS in the SPCAM was related to a weaker Hadley cell owing to less southern hemisphere subtropical low cloud and hence shortwave reflection. A wider and more northward ITCZ in SPCAM caused an increase in absolute vorticity, promoting TC genesis. There were also, stronger convective coupled equatorial wave activities in SPCAM than in CPCAM; better-simulated ER-waves appeared to induce a higher number of TC genesis events in the SPCAM environment. In the future, more sophisticated multiscale modelling frameworks can be employed to improve the simulation of extreme events in relation to the large-scale environment, while achieving better computational efficiency.

AGU24 - Nonlinear Geophysics (NG)Mon, Dec 09, 2024, Washington DC & OnlineSession NG01-01: Bayesian posterior sampling in non-linear seismic inverse problem using deep generative priorAn inverse problem involves deducing the causes from observed effects within a system, often requiring the solution of partial differential equations that describe the system underlying physics. In geophysical seismic imaging, the goal is to reconstruct subsurface structures (typically velocity and density fields) by analyzing seismic waveforms recorded at the Earth's surface. This involves solving the non-linear wave equation to model the propagation of seismic waves through the Earth. Full Waveform Inversion (FWI) is a deterministic technique using gradient methods. However, FWI faces challenges such as non-uniqueness and computational complexity, emphasizing the need for advanced methods to accurately quantify uncertainties in inferred subsurface properties.Bayesian inference offers a robust framework for solving inverse problems and quantifying uncertainties by applying Bayes' theorem. This approach constructs a posterior probability density function of model parameters given the observed data. In this work, we introduce the approach that parameterizes the unknowns by training Wasserstein Generative Adversarial Networks (WGAN), to learn the prior distribution in the latent space capable of generating realistic subsurface images. Once trained, the GAN remains fixed, providing a powerful generative prior for Bayesian posterior sampling.In this work, we experiment and compare various posterior sampling methods, including Markov chain Monte Carlo (McMC) sampling, variational Bayesian inference methods using normalizing flow (NF), inference neural network (INN), Stein Variational Gradient Descent (SVGD). We evaluate their performance in terms of computational efficiency and accuracy in capturing the posterior distribution. The integration of deep generative priors with various Bayesian sampling techniques demonstrates significant improvements in handling the high-dimensionality of non-linear geophysical inverse problems, offering a direction for future research in uncertainty quantification.

and Forecasting (ERF) code is a next-generation atmospheric flow solver that can leverage modern computational resources. • ERF contains key capabilities that enable the simulation of atmospheric flows relevant to atmospheric science and wind energy technology. • ERF has been verified against benchmark data for a variety of idealized test cases.

Stets, E.G.1*, M.J. Cashman1, O. Miller2, K.A. Powlen3, A. Martinez1, J. Padilla1, and A. A. Archer1* - Corresponding author (estets@usgs.gov)1 – U.S. Geological Survey, Water Resources Mission Area2 – U.S. Geological Survey, Utah Water Science Center3 – U.S. Geological Survey, Oklahoma-Texas Water Science Center

This study evaluates the impact of crown density on Soil Organic Carbon (SOC) in Shivapuri Nagarjun National Park (SNNP), Nepal. SOC levels vary across forest types due to differences in tree species, soil, and topographic factors. The research quantifies carbon stocks and develops predictive models incorporating crown density as key variables. While trees are known to enrich soil carbon, the specific role of crown density in SOC dynamics remains insufficiently understood, requiring further research for effective management. In this study, a total of 108 plots were randomly selected using stratified random sampling, extracting ~400 grams of soil with 5.7 cm diameter, 10 cm height soil corer. Samples were collected in zipped polythene bags and analyzed in the laboratory. Field measurements followed FRA (2015) guidelines, and SOC concentrations were determined using the Walkley-Black wet oxidation method (1934). Data analysis in R included ANOVA tests for model validation. Total SOC stock was 219.403 t/ha across four forest types in SNNP, with a significant positive correlation between SOC and crown density (p = 2.2*10⁻¹⁶). Among nine models tested, Model 5 (SOC_Stock = A + B * CrownDensity²) was the best fit, with higher adj. R² and lower RMSE. Results showed crown density significantly influences SOC stock across all forest types. These findings provide valuable insights to the local communities, park authorities, and researchers by enhancing understanding of soil and vegetation characteristics. Forest management should also consider tree variables like height, diameter, and stand age to optimize carbon stock assessment and management.

 Serpentinisation results from hydration wt. 5% H 2 O of a Phanerozoic lherzolites  High-temperature serpentinisation that occurred in the Mesozoic produced a large amount of H 2  Current H 2 emanation in the Pyrenees suggests that low-temperature serpentinization processes may be at play