Search NASA⌕ Search

Engineering topics

Collins, William D

Publications and source records attributed to Collins, William D.

Huge ensembles – Part 1: Design of ensemble weather forecasts using spherical Fourier neural operators

Abstract. Simulating low-likelihood high-impact extreme weather events in a warming world is a significant and challenging task for current ensemble forecasting systems. While these systems presently use up to 100 members, larger ensembles could enrich the sampling of internal variability. They may capture the long tails associated with climate hazards better than traditional ensemble sizes. Due to computational constraints, it is infeasible to generate huge ensembles (comprised of 1000–10 000 members) with traditional, physics-based numerical models. In this two-part paper, we replace traditional numerical simulations with machine learning (ML) to generate hindcasts of huge ensembles. In Part 1, we construct an ensemble weather forecasting system based on spherical Fourier neural operators (SFNOs), and we discuss important design decisions for constructing such an ensemble. The ensemble represents model uncertainty through perturbed-parameter techniques, and it represents initial condition uncertainty through bred vectors, which sample the fastest-growing modes of the forecast. Using the European Centre for Medium-Range Weather Forecasts Integrated Forecasting System (IFS) as a baseline, we develop an evaluation pipeline composed of mean, spectral, and extreme diagnostics. With large-scale, distributed SFNOs with 1.1 billion learned parameters, we achieve calibrated probabilistic forecasts. As the trajectories of the individual members diverge, the ML ensemble mean spectra degrade with lead time, consistent with physical expectations. However, the individual ensemble members' spectra stay constant with lead time. Therefore, these members simulate realistic weather states during the rollout, and the ML ensemble passes a crucial spectral test in the literature. The IFS and ML ensembles have similar extreme forecast indices, and we show that the ML extreme weather forecasts are reliable and discriminating. These diagnostics ensure that the ensemble can reliably simulate the time evolution of the atmosphere, including low-likelihood high-impact extremes. In Part 2, we generate a huge ensemble initialized each day in summer 2023, and we characterize the simulations of extremes.

Mahesh, Ankur↗

Huge ensembles – Part 2: Properties of a huge ensemble of hindcasts generated with spherical Fourier neural operators

Abstract. In Part 1, we created an ensemble based on spherical Fourier neural operators. As initial condition perturbations, we used bred vectors, and as model perturbations, we used multiple checkpoints trained independently from scratch. Based on diagnostics that assess the ensemble's physical fidelity, our ensemble has comparable performance to operational weather forecasting systems. However, it requires orders-of-magnitude fewer computational resources. Here in Part 2, we generate a huge ensemble (HENS), with 7424 members initialized each day of summer 2023. We enumerate the technical requirements for running huge ensembles at this scale. HENS precisely samples the tails of the forecast distribution and presents a detailed sampling of internal variability. HENS has two primary applications: (1) as a large dataset with which to study the statistics and drivers of extreme weather and (2) as a weather forecasting system. For extreme climate statistics, HENS samples events 4σ away from the ensemble mean. At each grid cell, HENS increases the skill of the most accurate ensemble member and enhances coverage of possible future trajectories. As a weather forecasting model, HENS issues extreme weather forecasts with better uncertainty quantification. It also reduces the probability of outlier events, in which the verification value lies outside the ensemble forecast distribution.

Mahesh, Ankur↗

Progress in Fast, Accurate Multi-scale Climate Simulations

We present a survey of physical and computational techniques that have the potential to con- tribute to the next generation of high-fidelity, multi-scale climate simulations. Examples of the climate science problems that can be investigated with more depth include the capture of remote forcings of localized hydrological extreme events, an accurate representation of cloud features over a range of spatial and temporal scales, and parallel, large ensembles of simulations to more effectively explore model sensitivities and uncertainties. Numerical techniques, such as adaptive mesh refinement, implicit time integration, and separate treatment of fast physical time scales are enabling improved accuracy and fidelity in simulation of dynamics and allow more complete representations of climate features at the global scale. At the same time, part- nerships with computer science teams have focused on taking advantage of evolving computer architectures, such as many-core processors and GPUs, so that these approaches which were previously considered prohibitively costly have become both more efficient and scalable. In combination, progress in these three critical areas is poised to transform climate modeling in the coming decades.

climate modeling↗

A Hierarchical Evaluation of Regional Climate Simulations

Global climate models (GCMs) are the primary tools for predicting the evolution of the climate system. Through decades of development, GCMs have demonstrated useful skill in simulating climate at continental to global scales. However, large uncertainties remain in projecting climate change at regional scales, which limit our ability to inform decisions on climate change adaptation and mitigation. To bridge this gap, different modeling approaches including nested regional climate models (RCMs), global stretch-grid models, and global high-resolution atmospheric models have been used to provide regional climate simulations (Leung et al. 2003). In previous efforts to evaluate these approaches, isolating their relative merits was not possible because factors such as dynamical frameworks, physics parameterizations, and model resolutions were not systematically constrained. With advances in high performance computing, it is now feasible to run coupled atmosphere-ocean GCMs at horizontal resolution comparable to what RCMs use today. Global models with local refinement using unstructured grids have become available for modeling regional climate (e.g., Rauscher et al. 2012; Ringler et al. 2013). While they offer opportunities to improve climate simulations, significant efforts are needed to test their veracity for regional-scale climate simulations.

climate↗

Toward a Minimal Representation of Aerosols in Climate Models: Description and Evaluation in the Community Atmosphere Model CAM5

A modal aerosol module (MAM) has been developed for the Community Atmosphere Model version 5 (CAM5), the atmospheric component of the Community Earth System Model version 1 (CESM1). MAM is capable of simulating the aerosol size distribution and both internal and external mixing between aerosol components, treating numerous complicated aerosol processes and aerosol physical, chemical and optical properties in a physically based manner. Two MAM versions were developed: a more complete version with seven-lognormal modes (MAM7), and a three-lognormal mode version (MAM3) for the purpose of long-term (decades to centuries) simulations. Major approximations in MAM3 include assuming immediate mixing of primary organic matter (POM) and black carbon (BC) with other aerosol components, merging of the MAM7 fine dust and fine sea salt modes into the accumulation mode, merging of the MAM7 coarse dust and coarse sea salt modes into the single coarse mode, and neglecting the explicit treatment of ammonia and ammonium cycles. Simulated sulfate and secondary organic aerosol (SOA) mass concentrations are remarkably similar between MAM3 and MAM7 as most ({approx}90%) of these aerosol species are in the accumulation mode. Differences of POM and BC concentrations between MAM3 and MAM7 are also small (mostly within 10%) because of the assumed hygroscopic nature of POM, so that freshly emitted POM and BC are wet-removed before mixing internally with soluble aerosol species. Sensitivity tests with the POM assumed to be hydrophobic and with slower aging process increase the POM and BC concentrations, especially at high latitudes (by several times). The mineral dust global burden differs by 10% and sea salt burden by 30-40% between MAM3 and MAM7 mainly due to the different size ranges for dust and sea salt modes and different standard deviations of log-normal size distribution for sea salt modes between MAM3 and MAM7. The model is able to qualitatively capture the observed geographical and temporal variations of aerosol mass and number concentrations, size distributions, and aerosol optical properties. However, there are noticeable biases, e.g., simulated sulfate and mineral dust concentrations at surface over the oceans are too low. Simulated BC concentrations are significant low in the Arctic. There is a low bias in modeled aerosol optical depth on the global scale, especially in the developing counties. There biases in aerosol simulations clearly indicate the need for improvements of aerosol processes (e.g., emission fluxes of anthropogenic aerosols and precursor gases in developing countries, boundary layer nucleation) and properties (e.g., primary aerosol emission size, POM hygroscopicity). In addition the critical role of cloud properties (e.g., liquid water content, cloud fraction) responsible for the wet scavenging of aerosol is highlighted.

54 ENVIRONMENTAL SCIENCES↗

Application of the CALIOP Layer Product to Evaluate the Vertical Distribution of Aerosols Estimated by Global Models: AeroCom Phase I Results

The CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) layer product is used for a multimodel evaluation of the vertical distribution of aerosols. Annual and seasonal aerosol extinction profiles are analyzed over 13 sub-continental regions representative of industrial, dust, and biomass burning pollution, from CALIOP 2007-2009 observations and from AeroCom (Aerosol Comparisons between Observations and Models) 2000 simulations. An extinction mean height diagnostic (Z{sub a}) is defined to quantitatively assess the models performance. It is calculated over the 0-6 km and 0-10 km altitude ranges by weighting the altitude of each 100 m altitude layer by its aerosol extinction coefficient. The mean extinction profiles derived from CALIOP layer products provide consistent regional and seasonal specificities and a low inter-annual variability. While the outputs from most models are significantly correlated with the observed Z{sub a} climatologies, some do better than others, and 2 of the 12 models perform particularly well in all seasons. Over industrial and maritime regions, most models show higher Z{sub a} than observed by CALIOP, whereas over the African and Chinese dust source regions, Z{sub a} is underestimated during Northern Hemisphere Spring and Summer. The positive model bias in Z{sub a} is mainly due to an overestimate of the extinction above 6 km. Potential CALIOP and model limitations, and methodological factors that might contribute to the differences are discussed.

09 BIOMASS FUELS↗

Using Surface Remote Sensors to Derive Radiative Characteristics of Mixed-Phase Clouds: An Example from M-PACE

Measurements from ground-based cloud radar, high spectral resolution lidar and microwave radiometer are used in conjunction with a column version of the Rapid Radiative Transfer Model (RRTMG) and radiosonde measurements to derive the surface radiative properties under mixed-phase cloud conditions. These clouds were observed during the United States Department of Energy (US DOE) Atmospheric Radiation Measurement (ARM) Mixed-Phase Arctic Clouds Experiment (M-PACE) between September and November of 2004. In total, sixteen half hour time periods are reviewed due to their coincidence with radiosonde launches. Cloud liquid (ice) water paths are found to range between 11.0-366.4 (0.5-114.1) gm-2, and cloud physical thicknesses fall between 286-2075 m. Combined with temperature and hydrometeor size estimates, this information is used to calculate surface radiative flux densities using RRTMG, which are demonstrated to generally agree with measured flux densities from surface-based radiometric instrumentation. Errors in longwave flux density estimates are found to be largest for thin clouds, while shortwave flux density errors are generally largest for thicker clouds. A sensitivity study is performed to understand the impact of retrieval assumptions and uncertainties on derived surface radiation estimates. Cloud radiative forcing is calculated for all profiles, illustrating longwave dominance during this time of year, with net cloud forcing generally between 50 and 90 Wm-2.

54 ENVIRONMENTAL SCIENCES↗