Joseph Gomes

This study presents a feature-scale cross-validation of the Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) cloud–aerosol discrimination (CAD) products using 9289 globally collocated Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) observations acquired between October 2018 and June 2023. The operational ATL09 and a U-Net convolutional neural network (CNN) products are evaluated against CALIPSO on a pixel-by-pixel basis using class-specific metrics. Agreement improves with decreasing along-track comparison-window width, which limits spatial divergence from the ground-track crossing, whereas agreement varies weakly and non-monotonically across the actual 0–10 min inter-satellite time separation. Agreement also depends strongly on the CALIPSO integration size used for layer detection, which affects both feature detectability and effective spatial resolution. Restricting the analysis to ±60° latitude markedly increases aerosol agreement, consistent with reduced influence from high-latitude ambiguities associated with blowing snow, diamond dust, and larger crossing angles, while cloud metrics change more modestly. At a 1-s comparison width and 5-km maximum CALIPSO integration scale, CNN–CALIPSO F1 exceeds ATL09–CALIPSO F1 by 0.027–0.045 globally, with all paired 95% confidence intervals above zero across cloud and aerosol classes under daytime and nighttime conditions; however, some differences within ±60° are not distinguishable from zero. These results show that agreement is governed not only by retrieval algorithms but also by differences in instrument characteristics, effective spatial resolution, and sampling strategy. The proposed collocation framework provides a basis for evaluating current and future machine-learning approaches for spaceborne lidar CAD.

Machine learning interatomic potentials (MLIPs) have surged in popularity over the last two decades, with many model architectures now openly available. As data-driven models, MLIPs critically depend on high-fidelity (i.e., physically accurate) training data produced by electronic structure calculations. However, assembling large and chemically diverse datasets can be a complex and time-consuming endeavor, often requiring the manual selection of representative atomic configurations and the execution of hundreds to millions of electronic structure simulations. To address this challenge, we introduce the Active Learning Framework (ALF), an open-source Python package designed to streamline the design and deployment of MLIP training datasets on High Performance Computing resources. ALF automatically selects new configurations from undersampled regions of the potential energy surface where the MLIP exhibits high uncertainty, schedules electronic structure calculations across available computational resources, and retrains MLIPs on the fly, thereby reducing manual intervention and limiting human bias. As a demonstration, we applied ALF to generate an actively learned dataset for molten salt mixtures consisting of F, Li, Na, Be, and K atoms. An MLIP trained on this data was then employed to predict melting point, viscosity, density, radial distribution function, and specific heat, which are computationally resource-intensive to evaluate via first-principles molecular dynamics. These results were subsequently validated against experimental data. Collectively, these findings illustrate ALF’s effectiveness in compiling datasets that capture essential chemical and structural regimes, thereby virtually eliminating manual curation.

Laser-based remote sensing (lidar) is a proven technique for detecting atmospheric features such as clouds and aerosols as well as for determining their vertical distribution with high accuracy. Even simple elastic backscatter lidars can distinguish clouds from aerosols, and accurate knowledge of their vertical location is essential for air quality assessment, hazard avoidance, and operational decision-making. However, daytime lidar measurements suffer from reduced signal-to-noise ratio (SNR) due to solar background contamination. Conventional processing approaches mitigate this by applying horizontal and vertical averaging, which improves SNR at the expense of spatial resolution and feature detectability. This work presents a deep learning-based framework that enhances lidar SNR at native resolution and performs fast layer detection and cloud–aerosol discrimination. We apply this approach to ICESat-2 532 nm photon-counting data, using artificially noised nighttime profiles to generate simulated daytime observations for training and evaluation. Relative to the simulated daytime data, our method improves peak SNR by more than a factor of three while preserving structural similarity with true nighttime profiles. After recalibration, the denoised photon counts yield an order-of-magnitude reduction in mean absolute percentage error in calibrated attenuated backscatter compared with the simulated daytime data, when validated against real nighttime measurements. We further apply the trained model to a full month of real daytime ICESat-2 observations (April 2023) and demonstrate effective layer detection and cloud–aerosol discrimination, maintaining high recall for both clouds and aerosols and showing qualitative improvement relative to the standard ATL09 data products. As an alternative to traditional averaging-based workflows, this deep learning approach offers accurate, near real-time data processing at native resolution. A key implication is the potential to enable smaller, lower-power spaceborne lidar systems that perform as well as larger instruments.

We present an improved method for determining guess structures for transition state searches by incorporating internal coordinate interpolation into the freezing string method (FSM). We test our method on over 40 reactions across 3 benchmark data sets covering a diverse set of chemical reactions. Our results show that incorporation of internal coordinate interpolation improves the reliability of the FSM, enabling larger interpolation step sizes and fewer optimization steps per cycle, which together yield nearly a 50% reduction in computational cost while maintaining a 100% success rate on benchmark chemical reaction test cases, including systems where previous attempts based on linear synchronous transit interpolation have failed. We provide an open-source Python implementation of the FSM, in addition to the reactant, product, and transition state structures of all reactions studied.

Space-based, photon-counting lidar instruments are effective tools for observing cloud and aerosol layers in the atmosphere. Cloud phases and several different kinds of aerosols are presently identified and typed using sophisticated, fine-tuned classification algorithms that operate using processed lidar data. We present a deep neural network semantic segmentation model that was trained using raw, uncalibrated photon count data and data products from the Cloud/Aerosol Transport System’s (CATS) 1064 nm lidar. Our approach successfully types layers in complex scenes using only raw photon counts, bin altitudes, and ground surface type at 14 to 171 times the spatial resolution of the CATS operational data product. We observe comparable cloud detection and phase determination to the CATS operational algorithm while also exhibiting a 15-point improvement in finding tenuous aerosol layers. Because the model is lightweight, does not rely upon ancillary information, and is optimized to leverage GPU computing, it has the potential to be deployed on-instrument to perform cloud and aerosol typing in real time.

This study presents a global aerosol climatology derived from six years (October 2018–October 2024) of the Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) observations, using a U-Net Convolutional Neural Network (CNN) machine learning algorithm for Cloud–Aerosol Discrimination (CAD). Despite ICESat-2’s design primarily as an altimetry mission with a single-wavelength, low-power, high-repetition-rate laser, ICESat-2 effectively captures global aerosol distribution patterns and can provide valuable insights to bridge the observational gap between the Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) and Earth Cloud, Aerosol and Radiation Explorer (EarthCARE) missions to support future spaceborne lidar mission design. The machine learning approach outperforms traditional thresholding methods, particularly in complex conditions of cloud embedded in aerosol, owing to a finer spatiotemporal resolution. Our results show that annually, between 60°S and 60°N, 78.4%, 17.0%, and 4.5% of aerosols are located within the 0–2 km, 2–4 km, and 4–6 km altitude ranges, respectively. Regional analyses cover the Arabian Sea (ARS), Arabian Peninsula (ARP), South Asia (SAS), East Asia (EAS), Southeast Asia (SEA), the Americas, and tropical oceans. Vertical aerosol structures reveal strong trans-Atlantic dust transport from the Sahara in summer and biomass burning smoke transport from the Savanna during dry seasons. Marine aerosol belts are most prominent in the tropics, contrasting with earlier reports of the Southern Ocean maxima. This work highlights the importance of vertical aerosol distributions needed for more accurate quantification of the aerosol–cloud interaction influence on radiative forcing for improving global climate models.

NASA’s Ice, Cloud, and land Elevation Satellite (ICESat-2), designed for surface altimetry, plays a pivotal role in providing precise ice sheet elevation measurements. While its primary focus is altimetry, ICESat-2 also offers valuable atmospheric data. Current conventional processing methods for producing atmospheric data products encounter challenges, particularly in conditions with low signal or high background noise. The thresholding technique traditionally used for atmospheric feature detection in lidar data uses a threshold value to accept signals while rejecting noise, which may result in signal loss or false detection in the presence of excessive noise. Traditional approaches for improving feature detection, such as averaging, lead to a trade-off between detection resolution and accuracy. In addition, the discrimination of cloud from aerosol in the identified features is difficult given ICESat-2’s single wavelength and lack of depolarization measurement capability. To address these challenges, we demonstrate atmospheric feature detection and cloud–aerosol discrimination using deep learning-based semantic segmentation by a convolutional neural network (CNN). The key findings from our research are the effectiveness of a deep learning model for feature detection and cloud–aerosol classification in ICESat-2 atmospheric data and the model’s surprising capability to detect complex atmospheric features at a finer resolution than is currently possible with traditional processing techniques. We identify several examples where the traditional feature detection and cloud–aerosol discrimination algorithms struggle, like in scenarios with several layers of vertically stacked clouds, or in the presence of clouds embedded within aerosol, and demonstrate the ability of the CNN model to detect such features, resolving the boundaries between adjacent layers and detecting clouds hidden within aerosol layers at a fine resolution.

Surface wettability plays an important role in heterogeneous electrocatalysis. Here we report a facile laser ablation strategy to directly modify the wettability of the silver catalyst surface and investigate its effect on oxygen reduction reaction (ORR). A broad range tuning of 2e⁻/4e⁻ ORR pathways was achieved, with hydrophilic silver surfaces (contact angle (θ_w) 31.1° ± 0.6°) showing high activity and selectivity towards 4e⁻ reduction of oxygen to water.

Many applications of machine learning require a model to make accurate pre-dictions on test examples that are distributionally different from training ones, while task-specific labels are scarce during training. An effective approach to this challenge is to pre-train a model on related tasks where data is abundant, and then fine-tune it on a downstream task of interest. While pre-training has been effective in many language and vision domains, it remains an open question how to effectively use pre-training on graph datasets. In this paper, we develop a new strategy and self-supervised methods for pre-training Graph Neural Networks (GNNs). The key to the success of our strategy is to pre-train an expressive GNN at the level of individual nodes as well as entire graphs so that the GNN can learn useful local and global representations simultaneously. We systematically study pre-training on multiple graph classification datasets. We find that naive strategies, which pre-train GNNs at the level of either entire graphs or individual nodes, give limited improvement and can even lead to negative transfer on many downstream tasks. In contrast, our strategy avoids negative transfer and improves generalization significantly across downstream tasks, leading up to 9.4% absolute improvements in ROC-AUC over non-pre-trained models and achieving state-of-the-art performance for molecular property prediction and protein function prediction.

The classical simulation of quantum systems typically requires exponential resources. Recently, the introduction of a machine learning-based wavefunction ansatz has led to the ability to solve the quantum many-body problem in regimes that had previously been intractable for existing exact numerical methods. Here, we demonstrate the utility of the variational representation of quantum states based on artificial neural networks for performing quantum optimization. We show empirically that this methodology achieves high approximation ratio solutions with polynomial classical computing resources for a range of instances of the Maximum Cut (MaxCut) problem whose solutions have been encoded into the ground state of quantum many-body systems up to and including 256 qubits.