Search

Economic growth, air pollution, and forest fires in some states in the United States have increased the concentration of particulate matter with a diameter less than or equal to 2.5 μm (PM2.5). Although previous studies have tried to observe PM2.5 both spatially and temporally using aerosol remote sensing and geostatistical estimation, they were limited in accuracy by coarse resolution. In this paper, the performance of machine learning models on predicting PM2.5 is assessed with linear regression (LR), decision tree (DT), gradient boosting regression (GBR), AdaBoost regression (ABR), XGBoost (XGB), k‐nearest neighbors (K‐NN), long short‐term memory (LSTM), random forest (RF), and support vector machine (SVM) using PM2.5 station data from 2017 to 2021. To compare the accuracy of all the nine machine learning models, the coefficient of determination (R2), root mean square error (RMSE), Nash‐Sutcliffe efficiency (NSE), root mean square error ratio (RSR), and percent bias (PBIAS) were evaluated. Among all nine models, the RF (100 decision trees with a max depth of 20) and support vector regression (SVR; nonlinear kernel, degree 3 polynomial) models were the best for predicting PM2.5 concentrations. Additionally, comparison of the PM2.5 performance metrics displayed that the models had better predictive behavior in the western United States than that in the eastern United States. Key Points: We present the prediction of PM2.5 concentrations over the United States using various machine learning (ML) algorithmsWe show ML as a new approach for analyzing large data sets due to the computational speed and easy implementation for massive amounts of dataThe study is important for improving our understanding of the differences among ML algorithms for Earth Science research [ABSTRACT FROM AUTHOR]

Classifying images using supervised machine learning (ML) relies on labeled training data—classes or text descriptions, for example, associated with each image. Data‐driven models are only as good as the data used for training, and this points to the importance of high‐quality labeled data for developing a ML model that has predictive skill. Labeling data is typically a time‐consuming, manual process. Here, we investigate the process of labeling data, with a specific focus on coastal aerial imagery captured in the wake of hurricanes that affected the Atlantic and Gulf Coasts of the United States. The imagery data set is a rich observational record of storm impacts and coastal change, but the imagery requires labeling to render that information accessible. We created an online interface that served labelers a stream of images and a fixed set of questions. A total of 1,600 images were labeled by at least two or as many as seven coastal scientists. We used the resulting data set to investigate interrater agreement: the extent to which labelers labeled each image similarly. Interrater agreement scores, assessed with percent agreement and Krippendorff's alpha, are higher when the questions posed to labelers are relatively simple, when the labelers are provided with a user manual, and when images are smaller. Experiments in interrater agreement point toward the benefit of multiple labelers for understanding the uncertainty in labeling data for machine learning research. Plain Language Summary: After hurricanes and storms, pictures taken from a plane can be used to observe how the coast was impacted. A single flight might take thousands of pictures. If a computer could automatically analyze the pictures, then a person would not need to look at them one‐by‐one. To teach a computer to analyze images, we need many pictures and many labels that describe what is visible in each picture. But where do we get those labels? Typically, a coastal scientist labels the pictures by sorting them into folders or typing codes into a spreadsheet. But does every coastal scientist label pictures the same way? Some labeling questions are easy to answer, and scientists mostly agree ("Is this image all water?"). Other labeling questions are harder to answer, and cause disagreement ("Was there damage to buildings?"). This paper is about how well scientists agree when labeling the same pictures, and how we can improve agreement among scientists. We try some experiments and offer a few ideas on how to improve agreement. We suggest writing very clear questions, using smaller images, and having a comprehensive manual. It turns out that having a manual with examples—and reading the manual!—really helps. Key Points: We measure agreement among coastal scientists labeling the same sets of poststorm imagesCoastal scientists agree more when rating landforms, less when labeling inferred processesIterating on questions, providing documentation, and using smaller image sizes all increase agreement [ABSTRACT FROM AUTHOR]

Sentiment analysis, a form of data analytics, utilises information from online discussions, reviews, and social media posts to assess consumer sentiments. This study utilised data collected from social media using online listening procedures to assess online sentiments on oysters from January 2019 through December 2022. The analysis utilises machine learning algorithms to extract consumer sentiments, opinions, and demands from online chatter from different online domains. The online sentiments are determined as positive, negative, or neutral based on their word choice, tone, and context. The information provided gives insights into perception, which is valuable information for oyster producers, seafood industry stakeholders, and marketers to identify consumer preferences and formulate appropriate strategies accordingly. The results suggest that while farmed oysters are gaining popularity, there are still some concerns and criticisms around the industry. Positive words associated with mentions of oysters in general include 'great', 'love', 'delicious', 'enjoyed', and 'oyster bar', while negative words associated with oysters include 'water', 'raw oyster', 'bad', and 'not eat'. The overall percentage net sentiment associated with all oysters in the United States is positive at 63%. The net sentiment associated with wild oysters is positive, at 51%, and that of farmed oysters is 58%. The oyster industry could invest more in public education, sustainability, and water‐cleaning initiatives to improve its image. Utilising social media to monitor and shape public perception can help the industry address concerns and enhance oyster‐related sentiments, offering valuable insights for marketing and sales strategies. [ABSTRACT FROM AUTHOR]

Modeling ecological patterns and processes often involve large‐scale and complex high‐dimensional spatial data. Due to the nonlinearity and multicollinearity of ecological data, traditional geostatistical methods have faced great challenges in model accuracy. As machine learning has increased our ability to construct models on big data, the main focus of the study is to propose the use of statistical models that hybridize machine learning and spatial interpolation methods to cope with increasingly large‐scale and complex ecological data. Here, two machine learning algorithms, boosted regression tree (BRT) and least absolute shrinkage and selection operator (LASSO), were combined with ordinary kriging (OK) to model plant invasions across the eastern United States. The accuracies of the hybrid models and conventional models were evaluated by 10‐fold cross‐validation. Based on an invasive plants dataset of 15 ecoregions across the eastern United States, the results showed that the hybrid algorithms were significantly better at predicting plant invasion when compared to commonly used algorithms in terms of RMSE and paired‐samples t‐test (with the p‐value <.0001). Besides, the additional aspect of the combined algorithms is to have the ability to select influential variables associated with the establishment of invasive cover, which cannot be achieved by conventional geostatistics. Higher accuracy in the prediction of large‐scale biological invasions improves our understanding of the ecological conditions that lead to the establishment and spread of plants into novel habitats across spatial scales. The results demonstrate the effectiveness and robustness of the hybrid BRTOK and LASOK that can be used to analyze large‐scale and high‐dimensional spatial datasets, and it has offered an optional source of models for spatial interpolation of ecology properties. It will also provide a better basis for management decisions in early‐detection modeling of invasive species. [ABSTRACT FROM AUTHOR]

Objectives: This study aimed to analyze publications on artificial intelligence (AI) for falls in older people from a bibliometric perspective. Methods: The Web of Science database was searched for titles of English‐language articles containing the words "artificial intelligence," "deep learning," "machine learning," "natural language processing,", "neural artificial network," "fall," "geriatric," "elderly," "aging," "older," and "old age." An R‐based application (Biblioshiny for bibliometrics) and VOSviewer software were used for analysis. Results: Thirty‐seven English articles published between 2018 and 2024 were included. The year 2023 is the year with the most publications with 16 articles. The most productive research field was "Engineering Electrical Electronic" with seven articles. The most productive country was the United States, followed by China. The most common words were "injuries," "people," and "risk factors." Conclusion: Publications on AI and falls in the elderly are both few in number and the number of publications has increased in recent years. Future research should include relevant analyses in scientific databases, such as Scopus and PubMed. [ABSTRACT FROM AUTHOR]

To better understand the regional changes in summertime temperatures across the conterminous United States (CONUS), we adopt a recently developed machine learning framework that can be used to reveal the timing of emergence of forced climate signals from the noise of internal climate variability. Specifically, we train an artificial neural network (ANN) on seasonally averaged temperatures across the CONUS and then task the ANN to output the year associated with an individual map. In order to correctly identify the year, the ANN must therefore learn time‐evolving patterns of climate change amidst the noise of internal climate variability. The ANNs are first trained and tested on data from large ensembles and then evaluated using observations from a station‐based data set. To understand how the ANN is making its predictions, we leverage a collection of ad hoc feature attribution methods from explainable artificial intelligence (XAI). We find that anthropogenic signals in seasonal mean minimum temperature have emerged by the early 2000s for the CONUS, which occurred earliest in the Eastern United States. While our observational timing of emergence estimates are not as sensitive to the spatial resolution of the training data, we find a notable improvement in ANN skill using a higher resolution climate model, especially for its early twentieth century predictions. Composites of XAI maps reveal that this improvement is linked to temperatures around higher topography. We find that increases in spatial resolution of the ANN training data may yield benefits for machine learning applications in climate science. Plain Language Summary: While temperatures around the world continue to warm due to human‐caused climate change, some areas have observed smaller temperature trends than others. Understanding this regional variability in the rate of warming is important when assessing future projections. One location that has observed less warming is across the United States during their summer season. To evaluate temperature variability in this region using real‐world observations and climate model simulations, we use a statistical method from artificial intelligence called neural networks. The goal of the neural network setup is to learn temperature patterns across the United States and then identify whether climate change effects have exceeded the range of natural variability that has occurred in the past. This is called the timing of emergence (ToE), which is the first year that the effect has clearly appeared. We find that the average United States minimum temperature increase has already emerged in historical records. However, we find no ToE for the average maximum temperature, other than in the Western United States. Another important finding of this study is that by using higher resolution climate model data (i.e., more latitude and longitude points), we find better accuracy in the neural network predictions. Key Points: Forced temperature changes have emerged in observations during summer in the United States as detected by an artificial neural networkIncreasing spatial resolution improves neural network skill for predicting the year of a given summer temperature mapWestern United States land surface climate properties contribute to earlier timing of emergence predictions for the SPEAR climate model [ABSTRACT FROM AUTHOR]

Data‐driven methods have been extensively applied to predict atmospheric compositions. Here, we explore the capability of a deep learning (DL) model to make ozone (O3) predictions across continents in China, the United States (US) and Europe. The DL model was trained and validated with surface O3 observations in China and the US in 2015–2018. The DL model was applied to predict hourly surface O3 over three continents in 2015–2022. Compared to baseline simulations using GEOS‐Chem (GC) model, our analysis exhibits mean biases of 2.6 and 4.8 μg/m3 with correlation coefficients of 0.94 and 0.93 (DL); and mean biases of 3.7 and 5.4 μg/m3 with correlation coefficients of 0.95 and 0.92 (GC) in Europe in 2015–2018 and 2019–2022, respectively. The comparable performances between DL and GC indicate the potential of DL to make reliable predictions over spatial and temporal domains where a wealth of local observations for training is not available. Plain Language Summary: Machine learning techniques have been extensively applied in the field of atmospheric science. It provides an efficient way of integrating data and predicting atmospheric compositions. Here, we explore the capability of a deep learning (DL) model to make ozone (O3) predictions across continents in China, the United States (US) and Europe. The DL model was trained and validated with surface O3 observations in China and the US in 2015–2018. We then applied the DL model to predict surface O3 concentrations in China, the US and Europe in 2015–2022. Our analysis exhibits comparable performances between DL and chemical transport models for surface O3 predictions in Europe. This indicates the potential of DL models to extend predictions across spatial and temporal domains. Key Points: Deep learning exhibits acceptable capabilities of spatial and temporal extrapolations for surface O3 predictionsComparable performances between deep learning and GEOS‐Chem models with respect to independent O3 observationsGood performance of deep learning for rapid hourly O3 predictions [ABSTRACT FROM AUTHOR]

Despite declines in total cardiovascular mortality rates in the United States, heart failure (HF) mortality rates as well as hospitalizations and readmissions have increased in the past decade. Increases have been relatively higher among young and middle‐aged adults (<65 years). Therefore, identification of individuals HF at‐risk (Stage A) or with pre‐HF (Stage B) before the onset of overt clinical signs and symptoms (Stage C) is urgently needed. Multivariate risk models (e.g., Pooled Cohort Equations to Prevent Heart Failure [PCP‐HF]) have been externally validated in diverse populations and endorsed by the 2022 HF Guidelines to apply a risk‐based framework for the prevention of HF. However, traditional risk factors included in the PCP‐HF model only account for half of an individual's lifetime risk of HF; novel risk factors (e.g., adverse pregnancy outcomes, impaired lung health, COVID‐19) are emerging as important risk‐enhancing factors that need to be accounted for in personalized approaches to prevention. In addition to determining the role of novel risk‐enhancing factors, integration of social determinants of health (SDoH) in identifying and addressing HF risk is needed to transform the current clinical paradigm for the prevention of HF. Comprehensive strategies to prevent the progression of HF must incorporate pharmacotherapies (e.g., sodium glucose co‐transporter‐2 inhibitors that have also been termed the "statins" of HF prevention), intensive blood pressure lowering, and heart‐healthy behaviors. Future directions include investigation of novel prediction models leveraging machine learning, integration of risk‐enhancing factors and SDoH, and equitable approaches to interventions for risk‐based prevention of HF. [ABSTRACT FROM AUTHOR]

Objective: Several studies of nurse staffing and patient outcomes found a curvilinear or U‐shaped relationship, with benefits from additional nurse staffing diminishing or reversing at high staffing levels. This study examined potential diminishing returns to nurse staffing and the existence of a "tipping point" or the level of staffing after which higher nurse staffing no longer improves and may worsen readmissions. Data Sources/Study Setting: The Readiness Evaluation And Discharge Interventions (READI) study database of over 130,000 adult (18+) inpatient discharges from 62 medical, surgical, and medical‐surgical (noncritical care) units from 31 United States (US) hospitals during October 2014–March 2017. Study Design: Observational cross‐sectional study using a fully nonparametric random forest machine learning method. Primary exposure was nurse hours per patient day (HPPD) broken down by registered nurses (nonovertime and overtime) and nonlicensed nursing personnel. The outcome was 30‐day all‐cause same‐hospital readmission. Partial dependence plots were used to visualize the pattern of predicted patient readmission risk along a range of unit staffing levels, holding all other patient characteristics and hospital and unit structural variables constant. Data Collection/Extraction methods: Secondary analysis of the READI data. Missing values were imputed using the missing forest algorithm in R. Principal Findings: Partial dependence plots were U‐shaped, showing the readmission risk first declining and then rising with additional nurse staffing. The tipping points were at 6.95 and 0.21 HPPD for registered nurse staffing (nonovertime and overtime, respectively) and 2.91 HPPD of nonlicensed nursing personnel. Conclusions: The U‐shaped association was consistent with diminishing returns to nurse staffing suggesting that incremental gains in readmission reduction from additional nurse staffing taper off and could reverse at high staffing levels. If confirmed in future causal analyses across multiple outcomes, accompanying investments in infrastructure and human resources may be needed to maximize nursing performance outcomes at higher levels of nurse staffing. [ABSTRACT FROM AUTHOR]

[The Geostationary Lightning Mapper (GLM) detects total lightning continuously from space, and does not distinguish intra‐cloud (IC) from cloud‐to‐ground (CG) lightning. This research focuses on differentiating CG and IC lightning detected by GLM using a random forests (RF) model. GLM flash and group characteristics are implemented into the RF model and used to predict flash type. The most important flash characteristic for distinguishing flash type is the maximum group area. Other features with high feature importance include time‐of‐day, elongation, propagation, footprint, slope, maximum distance between groups and events, and mean energy. Skill scores showcase the model's ability to distinguish flash type with moderate skill, with 81% probability of detection, 71% percent correct, 36% false alarm rate, 36% false alarm ratio, and 56% critical success index. These scores improve further when study area is limited to CONUS and the Southeastern United States. These results can be used to aid in future climatological analysis of flash type.] Plain Language Summary: The Geostationary Lightning Mapper (GLM) is a satellite‐based lightning sensor that allows for continuous detection of light emitted from the cloud tops due to lightning. GLM does not distinguish if the lightning is connecting to ground (CG) or remaining in the cloud (IC). In order to distinguish CG and IC flashes, this research uses a machine learning method called random forests (RF). The RF model attempts to classify lightning flashes based on their size, duration, and intensity. The most important flash characteristics for distinguishing flash type are the features related to the areal size of the lightning and the time of day the lightning occurs. Overall, moderate success is shown when attempting to divide total lightning into CG and IC over the 2018 period. This information can be used by researchers to improve the use of GLM in the study of different storm types as well as aiding in wildfire forecasting. Key Points: Although there is substantial overlap in the distributions for cloud‐to‐ground (CG) and intra‐cloud (IC) flash characteristics, the mean values differMost important features for flash distinction are maximum group area, time‐of‐day, elongation, propagation, footprint, slope, and max group/event distanceResults show skill in the random forests model's ability to distinguish CG and IC flashes in the Geostationary Lightning Mapper data [ABSTRACT FROM AUTHOR]

The fidelity of climate projections is often undermined by biases in climate models due to their simplification or misrepresentation of unresolved climate processes. While various bias correction methods have been developed to post‐process model outputs to match observations, existing approaches usually focus on limited, low‐order statistics, or break either the spatiotemporal consistency of the target variable, or its dependency upon model resolved dynamics. We develop a Regularized Adversarial Domain Adaptation (RADA) methodology to overcome these deficiencies, and enhance efficient identification and correction of climate model biases. Instead of pre‐assuming the spatiotemporal characteristics of model biases, we apply discriminative neural networks to distinguish historical climate simulation samples and observation samples. The evidences based on which the discriminative neural networks make distinctions are applied to train the domain adaptation neural networks to bias correct climate simulations. We regularize the domain adaptation neural networks using cycle‐consistent statistical and dynamical constraints. An application to daily precipitation projection over the contiguous United States shows that our methodology can correct all the considered moments of daily precipitation at approximately 1° resolution, ensures spatiotemporal consistency and inter‐field correlations, and can discriminate between different dynamical conditions. Our methodology offers a powerful tool for disentangling model parameterization biases from their interactions with the chaotic evolution of climate dynamics, opening a novel avenue toward big‐data enhanced climate predictions. Plain Language Summary: Accurate climate prediction is crucial for understanding climate change and implementing effective climate adaptation strategies. However, climate models that are used to generate climate predictions have multifaceted biases that often need to be corrected before predictions can be considered usable. We develop a data‐driven methodology that detects and corrects climate model biases using a game theory inspired machine learning technique. By applying physical and statistical constraints, our predictions are not only more accurate, but also more trustworthy as judged by our physical understandings. Key Points: Identify and correct climate projection biases using unpaired climate simulation and observation dataRegularize the data‐driven bias corrector using statistical and dynamical constraintsSignificant improvement of daily precipitation estimation regarding a broad range of spatiotemporal statistics [ABSTRACT FROM AUTHOR]

Understanding the complex interrelationships between wildfire and its environmental and anthropogenic controls is crucial for wildfire modeling and management. Although machine learning (ML) models have yielded significant improvements in wildfire predictions, their limited interpretability has been an obstacle for their use in advancing understanding of wildfires. This study builds an ML model incorporating predictors of local meteorology, land‐surface characteristics, and socioeconomic variables to predict monthly burned area at grid cells of 0.25° × 0.25° resolution over the contiguous United States. Besides these predictors, we construct and include predictors representing the large‐scale circulation patterns conducive to wildfires, which largely improves the temporal correlations in several regions by 14%–44%. The Shapley additive explanation is introduced to quantify the contributions of the predictors to burned area. Results show a key role of longitude and latitude in delineating fire regimes with different temporal patterns of burned area. The model captures the physical relationship between burned area and vapor pressure deficit, relative humidity (RH), and energy release component (ERC), in agreement with the prior findings. Aggregating the contribution of predictor variables of all the grids by region, analyses show that ERC is the major contributor accounting for 14%–27% to large burned areas in the western US. In contrast, there is no leading factor contributing to large burned areas in the eastern US, although large‐scale circulation patterns featuring less active upper‐level ridge‐trough and low RH two months earlier in winter contribute relatively more to large burned areas in spring in the southeastern US. Plain Language Summary: Wildfire burned areas have increased by 10 times since the 1980s in the United States, posing more threats to properties and human life, and degrading air quality. There is a demand for wildfire controls by accurate predictions and a better understanding of wildfires. Machine learning (ML) is an effective tool for resolving the nonlinear relationships between wildfire and its predictors. This study builds an ML model and incorporates a game‐theory‐based interpretation to predict wildfires and explain the relationships between burned area and their key controlling factors. We show that including predictors representing large‐scale meteorological patterns favorable for wildfires significantly improves burned area predictions. Using the novel interpretation method, we identify the roles of the coordinate variables in distinguishing fire regimes with different temporal patterns of burned area as well as the physical relationships between local meteorology and burned area. Additionally, we show that fuel dryness is the most important predictor of large burned areas in the western US while the large‐scale meteorological patterns featuring dry winters contribute more in the southeastern US. Our study provides a better elucidation of the complex processes contributing to wildfires using ML tools and the game theory interpretation. Key Points: US fire burned areas are well predicted by a machine learning model and SHAP improves interpretation of the contributing factorsIncluding large‐scale circulation patterns conducive to wildfires as predictors improve prediction of burned areas in several regionsFire‐season fuel dryness and dry winters are important contributors to large burned areas in western and southeastern US, respectively [ABSTRACT FROM AUTHOR]

Operational satellite remote sensing products are transforming rangeland management and science. Advancements in computation, data storage and processing have removed barriers that previously blocked or hindered the development and use of remote sensing products. When combined with local data and knowledge, remote sensing products can inform decision‐making at multiple scales.We used temporal convolutional networks to produce a fractional cover product that spans western United States rangelands. We trained the model with 52,012 on‐the‐ground vegetation plots to simultaneously predict fractional cover for annual forbs and grasses, perennial forbs and grasses, shrubs, trees, litter and bare ground. To assist interpretation and to provide a measure of prediction confidence, we also produced spatiotemporal‐explicit, pixel‐level estimates of uncertainty. We evaluated the model with 5,780 on‐the‐ground vegetation plots removed from the training data.Model evaluation averaged 6.3% mean absolute error and 9.6% root mean squared error. Evaluation with additional datasets that were not part of the training dataset, and that varied in geographic range, method of collection, scope and size, revealed similar metrics. Model performance increased across all functional groups compared to the previously produced fractional product.The advancements achieved with the new rangeland fractional cover product expand the management toolbox with improved predictions of fractional cover and pixel‐level uncertainty. The new product is available on the Rangeland Analysis Platform (https://rangelands.app/), an interactive web application that tracks rangeland vegetation through time. This product is intended to be used alongside local on‐the‐ground data, expert knowledge, land use history, scientific literature and other sources of information when making interpretations. When being used to inform decision‐making, remotely sensed products should be evaluated and utilized according to the context of the decision and not be used in isolation. [ABSTRACT FROM AUTHOR]

Over the past decade, remote sensing data have improved in resolution and become more widely available, bringing new opportunities for its use in environmental science and conservation. One potential application is to identify and map instream infrastructure across the world, with important implications for fisheries, hydrology, flooding, and more. To date, databases of instream infrastructure focus on larger dams with reservoirs that are comparatively easy to detect with remotely sensed imagery. Despite their impact on freshwater ecosystems, smaller infrastructure is often overlooked. To overcome these challenges, we require more systematic approaches, such as the Global River Obstruction Database (GROD) presented here, to map instream infrastructure. We present a participatory approach to identify, map, and validate infrastructure and provide an initial data set for the contiguous United States (n = 4,197). We highlight the value of participatory methods that include the public and suggest ways they could be fused with machine learning for future applications. Plain Language Summary: In recent years, imagery retrieved from Earth‐observing satellites has improved in quality and become more widely available. We can use the improved satellite imagery to observe the Earth's surface in entirely new ways. One potential application is to identify and map dams and other river obstructions that represent disturbances to the Earth's freshwater systems. Keeping record of river obstructions is valuable for many purposes from fisheries management to understanding flood dynamics. Currently, global dam databases focus on larger structures, while smaller dams and infrastructure are often overlooked. Small river obstructions greatly outnumber their larger counterparts and collectively have a significant impact on freshwater ecosystems. To further knowledge on the location and type of global river obstructions, we created the Global River Obstruction Database (GROD). GROD uses a uniform approach that maps river obstructions across the globe regardless of size or reservoir presence. We present and validate our citizen science‐based approach to map obstructions and provide an initial data set of 4,197 obstructions for the contiguous United States. We also highlight the value of incorporating members of the public in data collection and suggest ways citizen science‐based methods could be fused with more technical applications like machine learning for future projects. Key Points: We identify and map instream infrastructure in the contiguous United States as part of the Global River Obstructions DatabaseWe validate our approach against highly accurate regional data sets and find that we correctly identify a large fraction of infrastructureWe discuss how our participatory approach can be used with machine learning to further the mapping of global instream infrastructure [ABSTRACT FROM AUTHOR]

Premise: Apple orchards in the United States are under constant threat from a large number of pathogens and insects. Appropriate and timely deployment of disease management depends on early disease detection. Incorrect and delayed diagnosis can result in either excessive or inadequate use of chemicals, with increased production costs and increased environmental and health impacts. Methods and Results: We have manually captured 3651 high‐quality, real‐life symptom images of multiple apple foliar diseases, with variable illumination, angles, surfaces, and noise. A subset of images, expert‐annotated to create a pilot data set for apple scab, cedar apple rust, and healthy leaves, was made available to the Kaggle community for the Plant Pathology Challenge as part of the Fine‐Grained Visual Categorization (FGVC) workshop at the 2020 Computer Vision and Pattern Recognition conference (CVPR 2020). Participants were asked to use the image data set to train a machine learning model to classify disease categories and develop an algorithm for disease severity quantification. The top three area under the ROC curve (AUC) values submitted to the private leaderboard were 0.98445, 0.98182, and 0.98089. We also trained an off‐the‐shelf convolutional neural network on this data for disease classification and achieved 97% accuracy on a held‐out test set. Discussion: This data set will contribute toward development and deployment of machine learning–based automated plant disease classification algorithms to ultimately realize fast and accurate disease detection. We will continue to add images to the pilot data set for a larger, more comprehensive expert‐annotated data set for future Kaggle competitions and to explore more advanced methods for disease classification and quantification. [ABSTRACT FROM AUTHOR]

This study tests the utility of convolutional neural networks as a postprocessing framework for improving the National Center for Environmental Prediction's Global Forecast System's integrated vapor transport forecast field in the Eastern Pacific and western United States. Integrated vapor transport is the characteristic field of atmospheric rivers, which provide over 65% of yearly precipitation at some western U.S. locations. The method reduces full‐field root‐mean‐square error (RMSE) at forecast leads from 3 hr to seven days (9–17% reduction), while increasing correlation between observations and predictions (0.5–12% increase). This represents an approximately one‐ to two‐day lead time improvement in RMSE. Decomposing RMSE shows that random error and conditional biases are predominantly reduced. Systematic error is reduced up to five‐day forecast lead, but accounts for a smaller portion of RMSE. This work demonstrates convolutional neural networks potential to improve forecast skill out to seven days for precipitation events affecting the western United States. Plain Language Summary: Machine learning methods are data‐driven algorithms that improve by examining massive amounts of existing data. We explore the utility of a computer‐vision machine learning technique to reduce error in numerical weather forecasts of the characteristic field for atmospheric rivers (ARs). ARs are long narrow corridors of anomalous vapor transport capable of providing both beneficial and hazardous precipitation. Therefore, accurately forecasting AR events is extremely important from a water supply and flood protection standpoint. We show significant forecast improvements by applying machine learning postprocessing for lead times ranging from 3 hr to seven days, making the predictions more valuable to stakeholders affected by AR events. Key Points: The GFS forecast field of integrated vapor transport is used for a convolutional neural network‐based forecast postprocessing methodThe machine learning algorithm reduces the full‐field root‐mean‐square error and improves the correlation with ground truthAn error deconstruction shows that the dominant improvements come from the reduction of random error and conditional biases [ABSTRACT FROM AUTHOR]

Objective: To assess the issue of nonrepresentative sampling in Medicare Advantage (MA) risk adjustment.Data Sources: Medicare enrollment and claims data from 2008 to 2011.Data Extraction: Risk adjustment predictor variables were created from 2008 to 2010 Part A and B claims and the Medicare Beneficiary Summary File. Spending is based on 2009-2011 Part A and B, Durable Medical Equipment, and Home Health Agency claims files.Study Design: A propensity-score matched sample of Traditional Medicare (TM) beneficiaries who resembled MA enrollees was created. Risk adjustment formulas were estimated using multiple techniques, and performance was evaluated based on R2 , predictive ratios, and formula coefficients in the matched sample and a random sample of TM beneficiaries.Principal Findings: Matching improved balance on observables, but performance metrics were similar when comparing risk adjustment formula results fit on and evaluated in the matched sample versus fit on the random sample and evaluated in the matched sample.Conclusions: Fitting MA risk adjustment formulas on a random sample versus a matched sample yields little difference in MA plan payments. This does not rule out potential improvements via the matching method should reliable MA encounter data and additional variables become available for risk adjustment. [ABSTRACT FROM AUTHOR]

Objective: To propose nonparametric double robust machine learning in variable importance analyses of medical conditions for health spending.Data Sources: 2011-2012 Truven MarketScan database.Study Design: I evaluate how much more, on average, commercially insured enrollees with each of 26 of the most prevalent medical conditions cost per year after controlling for demographics and other medical conditions. This is accomplished within the nonparametric targeted learning framework, which incorporates ensemble machine learning. Previous literature studying the impact of medical conditions on health care spending has almost exclusively focused on parametric risk adjustment; thus, I compare my approach to parametric regression.Principal Findings: My results demonstrate that multiple sclerosis, congestive heart failure, severe cancers, major depression and bipolar disorders, and chronic hepatitis are the most costly medical conditions on average per individual. These findings differed from those obtained using parametric regression.Conclusions: The literature may be underestimating the spending contributions of several medical conditions, which is a potentially critical oversight. If current methods are not capturing the true incremental effect of medical conditions, undesirable incentives related to care may remain. Further work is needed to directly study these issues in the context of federal formulas. [ABSTRACT FROM AUTHOR]