Search for: All records

Artificial intelligence (AI) tools are designed to improve the efficacy and efficiency of data analysis and interpretation by the human decision maker. However, we know little about the optimal ways to present AI output to providers. This study used radiology image interpretation with AI-based decision support to explore the impact of different forms of AI output on reader performance. Readers included 5 experienced radiologists and 3 radiology residents reporting on a series of COVID chest x-ray images. Four different forms (1 word summarizing diagnoses (normal, mild, moderate, severe), probability graph, heatmap, heatmap plus probability graph) of AI outputs (plus no AI feedback) were evaluated. Results reveal that most decisions regarding presence/absence of COVID without AI were correct and overall remained unchanged across all types of AI outputs. Fewer than 1% of decisions that were changed as a function of seeing the AI output were negative (true positive to false negative or true negative to false positive) regarding presence/absence of COVID; and about 1% were positive (false negative to true positive, false positive to true negative). More complex output formats (e.g., heat map plus a probability graph) tend to increase reading time and the number of scans between the clinical image and the AI outputs as revealed through eyetracking. The key to the success of AI tools in medical imaging will be to incorporate the human into the overall process to optimize and synergize the human-computer dyad, since at least for the foreseeable future, the human is and will be the ultimate decision maker. Our results demonstrate that the form of the AI output is important as it can impact clinical decision making and efficiency. 

Abstract During the Seeded and Natural Orographic Wintertime clouds: the Idaho Experiment (SNOWIE) field campaign, cloud-top generating cells were frequently observed in the very high-resolution W-band airborne cloud radar data. This study examines multiple flight segments from three SNOWIE cases that exhibited cloud-top generating cells structures, focusing on the in situ measurements inside and outside these cells to characterize the microphysics of these cells. The observed generating cells in these three cases occurred in cloud tops of −15° to −30°C, with and without overlying cloud layers, but always with shallow layers of atmospheric instability observed at cloud top. The results also indicate that liquid water content, vertical velocity, and drizzle and ice crystal concentrations are greater inside the generating cells compared to the adjacent portions of the cloud. The generating cells were predominantly <500 m in horizontal width and frequently exhibited drizzle drops coexisting with ice. The particle imagery indicates that ice particle habits included plates, columns, and rimed and irregular crystals, likely formed via primary ice nucleation mechanisms. Understanding the sources of natural ice formation is important to understanding precipitation formation in winter orographic clouds, and is especially relevant for clouds that may be targeted for glaciogenic cloud seeding as well as to improve model representation of these clouds. Significance StatementThis study presents the characteristics of cloud-top generating cells in winter orographic clouds, and documents that fine-scale generating cells are ubiquitous in clouds over complex terrain in addition to having been observed in other types of clouds. The generating cells exhibited enhanced concentrations of larger drizzle and ice particles, which suggests the environments of these fine-scale features promote ice formation and growth. The source of ice formation in winter clouds is critical to understanding and modeling the precipitation formation process. Given the ubiquity of cloud-top generating cells in many types of clouds around the world, this study further motivates the need to investigate methods for representing subgrid-scale environments to improve ice formation in numerical models. 

Abstract Iron emissions from human activities, such as oil combustion and smelting, affect the Earth's climate and marine ecosystems. These emissions are difficult to quantify accurately due to a lack of observations, particularly in remote ocean regions. In this study, we used long‐term, near‐source observations in areas with a dominance of anthropogenic iron emissions in various parts of the world to better estimate the total amount of anthropogenic iron emissions. We also used a statistical source apportionment method to identify the anthropogenic components and their sub‐sources from bulk aerosol observations in the United States. We find that the estimates of anthropogenic iron emissions are within a factor of 3 in most regions compared to previous inventory estimates. Under‐ or overestimation varied by region and depended on the number of sites, interannual variability, and the statistical filter choice. Smelting‐related iron emissions are overestimated by a factor of 1.5 in East Asia compared to previous estimates. More long‐term iron observations and the consideration of the influence of dust and wildfires could help reduce the uncertainty in anthropogenic iron emissions estimates. 

Abstract. Aerosol particles are an important part of the Earth climate system, and their concentrations are spatially and temporally heterogeneous, as well as being variable in size and composition. Particles can interact with incoming solar radiation and outgoing longwave radiation, change cloud properties, affect photochemistry, impact surface air quality, change the albedo of snow and ice, and modulate carbon dioxide uptake by the land and ocean. High particulate matter concentrations at the surface represent an important public health hazard. There are substantial data sets describing aerosol particles in the literature or in public health databases, but they have not been compiled for easy use by the climate and air quality modeling community. Here, we present a new compilation of PM2.5 and PM10 surface observations, including measurements of aerosol composition, focusing on the spatial variability across different observational stations. Climate modelers are constantly looking for multiple independent lines of evidence to verify their models, and in situ surface concentration measurements, taken at the level of human settlement, present a valuable source of information about aerosols and their human impacts complementarily to the column averages or integrals often retrieved from satellites. We demonstrate a method for comparing the data sets to outputs from global climate models that are the basis for projections of future climate and large-scale aerosol transport patterns that influence local air quality. Annual trends and seasonal cycles are discussed briefly and are included in the compilation. Overall, most of the planet or even the land fraction does not have sufficient observations of surface concentrations – and, especially, particle composition – to characterize and understand the current distribution of particles. Climate models without ammonium nitrate aerosols omit ∼ 10 % of the globally averaged surface concentration of aerosol particles in both PM2.5 and PM10 size fractions, with up to 50 % of the surface concentrations not being included in some regions. In these regions, climate model aerosol forcing projections are likely to be incorrect as they do not include important trends in short-lived climate forcers. 

Abstract In Part II, two classes of vertical motions, fixed (associated with vertically propagating gravity waves tied to flow over topography) and transient (associated primarily with vertical wind shear and conditional instability within passing weather systems), were diagnosed over the Payette River basin of Idaho during the Seeded and Natural Orographic Wintertime Clouds: The Idaho Experiment (SNOWIE). This paper compares vertical motions retrieved from airborne Doppler radial velocity measurements with those from a 900-m-resolution model simulation to determine the impact of transient vertical motions on trajectories of ice particles initiated by airborne cloud seeding. An orographic forcing index, developed to compare vertical motion fields retrieved from the radar with the model, showed that fixed vertical motions were well resolved by the model while transient vertical motions were not. Particle trajectories were calculated for 75 cross-sectional pairs, each differing only by the observed and modeled vertical motion field. Wind fields and particle terminal velocities were otherwise identical in both trajectories so that the impact of transient vertical circulations on particle trajectories could be isolated. In 66.7% of flight-leg pairs, the distance traveled by particles in the model and observations differed by less than 5 km with transient features having minimal impact. In 9.3% of the pairs, model and observation trajectories landed within the ideal target seeding elevation range (>2000 m), whereas, in 77.3% of the pairs, both trajectories landed below the ideal target elevation. Particles in the observations and model descended into valleys on the mountains’ lee sides in 94.2% of cases in which particles traveled less than 37 km.