Decoding Physical and Cognitive Impacts of Particulate Matter Concentrations at Ultra-Fine Scales.

The human body is an incredible and complex sensing system. Environmental factors trigger a wide range of automatic neurophysiological responses. Biometric sensors can capture these responses in real time, providing clues about the underlying biophysical mechanisms. In this prototype study, we demonstrate an experimental paradigm to holistically capture and evaluate the interactions between an environmental context and physiological markers of an individual operating that environment. A cyclist equipped with a biometric sensing suite is followed by an environmental survey vehicle during outdoor bike rides. The interactions between environment and physiology are then evaluated though the development of empirical machine learning models, which estimate particulate matter concentrations from biometric variables alone. Here, we show biometric variables can be used to accurately estimate particulate matter concentrations at ultra-fine spatial scales with high fidelity (r2 = 0.91) and that smaller particles are better estimated than larger ones. Inferring environmental conditions solely from biometric measurements allows us to disentangle key interactions between the environment and the body. This work sets the stage for future investigations of these interactions for a larger number of factors, e.g., black carbon, CO2, NO/NO2/NOx, and ozone. By tapping into our body's 'built-in' sensing abilities, we can gain insights into how our environment influences our physical health and cognitive performance.

Data-Driven EEG Band Discovery with Decision Trees.

Electroencephalography (EEG) is a brain imaging technique in which electrodes are placed on the scalp. EEG signals are commonly decomposed into frequency bands called delta, theta, alpha, and beta. While these bands have been shown to be useful for characterizing various brain states, their utility as a one-size-fits-all analysis tool remains unclear. The goal of this work is to outline an objective strategy for discovering optimal EEG bands based on signal power spectra. A two-step data-driven methodology is presented for objectively determining the best EEG bands for a given dataset. First, a decision tree is used to estimate the optimal frequency band boundaries for reproducing the signal's power spectrum for a predetermined number of bands. The optimal number of bands is then determined using an Akaike Information Criterion (AIC)-inspired quality score that balances goodness-of-fit with a small band count. This data-driven approach led to better characterization of the underlying power spectrum by identifying bands that outperformed the more commonly used band boundaries by a factor of two. Additionally, key spectral components were isolated in dedicated frequency bands. The proposed method provides a fully automated and flexible approach to capturing key signal components and possibly discovering new indices of brain activity.

Machine Learning for Estimating Electron Transfer Rates From Square Wave Voltammetry.

Electrochemistry of surface-bound molecules is of high importance for numerous electronic and sensor applications. Extracting the electron transfer rate is beneficial for understanding surface-bound processes, but it requires experimental or computational rigor. We evaluate methods to determine electron transfer rates from large voltammetry sets from experiments via machine learning using decision tree ensembles, neural networks, and gaussian process regression models. We applied these to reproduce computational measures of electron transfer rates modeled by first principles. The best machine learning models were a random forest with 80 decision trees and a neural network with Bayesian regularization, producing root mean squared errors of 0.37 and 0.49 s-1 , respectively, corresponding to mean percent errors of 0.38 % and 0.52 %, respectively. This work establishes machine learning methods for rapidly acquiring electron transfer rates across large datasets for widespread applications.

Machine Learning for Light Sensor Calibration.

Sunlight incident on the Earth's atmosphere is essential for life, and it is the driving force of a host of photo-chemical and environmental processes, such as the radiative heating of the atmosphere. We report the description and application of a physical methodology relative to how an ensemble of very low-cost sensors (with a total cost of <$20, less than 0.5% of the cost of the reference sensor) can be used to provide wavelength resolved irradiance spectra with a resolution of 1 nm between 360-780 nm by calibrating against a reference sensor using machine learning. These low-cost sensor ensembles are calibrated using machine learning and can effectively reproduce the observations made by an NIST calibrated reference instrument (Konica Minolta CL-500A with a cost of around USD 6000). The correlation coefficient between the reference sensor and the calibrated low-cost sensor ensemble has been optimized to have R2> 0.99. Both the circuits used and the code have been made publicly available. By accurately calibrating the low-cost sensors, we are able to distribute a large number of low-cost sensors in a neighborhood scale area. It provides unprecedented spatial and temporal insights into the micro-scale variability of the wavelength resolved irradiance, which is relevant for air quality, environmental and agronomy applications.

Autonomous Learning of New Environments with a Robotic Team Employing Hyper-Spectral Remote Sensing, Comprehensive In-Situ Sensing and Machine Learning.

This paper describes and demonstrates an autonomous robotic team that can rapidly learn the characteristics of environments that it has never seen before. The flexible paradigm is easily scalable to multi-robot, multi-sensor autonomous teams, and it is relevant to satellite calibration/validation and the creation of new remote sensing data products. A case study is described for the rapid characterisation of the aquatic environment, over a period of just a few minutes we acquired thousands of training data points. This training data allowed for our machine learning algorithms to rapidly learn by example and provide wide area maps of the composition of the environment. Along side these larger autonomous robots two smaller robots that can be deployed by a single individual were also deployed (a walking robot and a robotic hover-board), observing significant small scale spatial variability.

Using Machine Learning for the Calibration of Airborne Particulate Sensors.

Airborne particulates are of particular significance for their human health impacts and their roles in both atmospheric radiative transfer and atmospheric chemistry. Observations of airborne particulates are typically made by environmental agencies using rather expensive instruments. Due to the expense of the instruments usually used by environment agencies, the number of sensors that can be deployed is limited. In this study we show that machine learning can be used to effectively calibrate lower cost optical particle counters. For this calibration it is critical that measurements of the atmospheric pressure, humidity, and temperature are also made.

Estimating the daily pollen concentration in the atmosphere using machine learning and NEXRAD weather radar data.

Millions of people have an allergic reaction to pollen. The impact of pollen allergies is on the rise due to increased pollen levels caused by global warming and the spread of highly invasive weeds. The production, release, and dispersal of pollen depend on the ambient weather conditions. The temperature, rainfall, humidity, cloud cover, and wind are known to affect the amount of pollen in the atmosphere. In the past, various regression techniques have been applied to estimate and forecast the daily pollen concentration in the atmosphere based on the weather conditions. In this research, machine learning methods were applied to the Next Generation Weather Radar (NEXRAD) data to estimate the daily Ambrosia pollen over a 300 km × 300 km region centered on a NEXRAD weather radar. The Neural Network and Random Forest machine learning methods have been employed to develop separate models to estimate Ambrosia pollen over the region. A feasible way of estimating the daily pollen concentration using only the NEXRAD radar data and machine learning methods would lay the foundation to forecast daily pollen at a fine spatial resolution nationally.

Using machine learning to understand the temporal morphology of the PM2.5 annual cycle in East Asia.

PM2.5 air pollution is a significant issue for human health all over the world, especially in East Asia. A large number of ground-based measurement sites have been established over the last decade to monitor real-time PM2.5 concentration. However, even this enhanced observational network leaves many gaps in characterizing the PM2.5 spatial distribution. Machine learning provides a variety of algorithms to help deal with these large spatial gaps-combining both remotely sensed and in situ observation data to estimate the global PM2.5 concentration. This study used a PM2.5 data product of six regions from the results of an unsupervised self-organizing map (SOM) with optimized ensemble learning approaches to highlight the most important meteorological and surface variables associated with PM2.5 concentration. These variables were then examined via multiple linear regression models to provide physical mechanistic insights into the morphology of the PM2.5 annual cycles.

Using machine learning to examine the relationship between asthma and absenteeism.

In this study, we found that machine learning was able to effectively estimate student learning outcomes geo-spatially across all the campuses in a large, urban, independent school district. The machine learning showed that key factors in estimating the student learning outcomes included the number of days students were absent from school. In turn, one of the most important factors in estimating the number of days a student was absent was whether or not the student had asthma. This highlights the importance of environmental public health for student learning outcomes.

Time-series analysis of satellite-derived fine particulate matter pollution and asthma morbidity in Jackson, MS.

In order to examine associations between asthma morbidity and local ambient air pollution in an area with relatively low levels of pollution, we conducted a time-series analysis of asthma hospital admissions and fine particulate matter pollution (PM2.5) in and around Jackson, MS, for the period 2003 to 2011. Daily patient-level records were obtained from the Mississippi State Department of Health (MSDH) Asthma Surveillance System. Patient geolocations were aggregated into a grid with 0.1° × 0.1° resolution within the Jackson Metropolitan Statistical Area. Daily PM2.5 concentrations were estimated via machine-learning algorithms with remotely sensed aerosol optical depth and other associated parameters as inputs. Controlling for long-term temporal trends and meteorology, we estimated a 7.2% (95% confidence interval 1.7-13.1%) increase in daily all-age asthma emergency room admissions per 10 µg/m3 increase in the 3-day average of PM2.5 levels (current day and two prior days). Stratified analyses reveal significant associations between asthma and 3-day average PM2.5 for males and blacks. Our results contribute to the current epidemiologic evidence on the association between acute ambient air pollution exposure and asthma morbidity, even in an area characterized by relatively good air quality.

Applying machine learning to forecast daily Ambrosia pollen using environmental and NEXRAD parameters.

Approximately 50 million Americans have allergic diseases. Airborne plant pollen is a significant trigger for several of these allergic diseases. Ambrosia (ragweed) is known for its abundant production of pollen and its potent allergic effect in North America. Hence, estimating and predicting the daily atmospheric concentration of pollen (ragweed pollen in particular) is useful for both people with allergies and for the health professionals who care for them. In this study, we show that a suite of variables including meteorological and land surface parameters, as well as next-generation radar (NEXRAD) measurements together with machine learning can be used to estimate successfully the daily pollen concentration. The supervised machine learning approaches we used included random forests, neural networks, and support vector machines. The performance of the training is independently validated using 10% of the data partitioned using the holdout cross-validation method from the original dataset. The random forests (R= 0.61, R2= 0.37), support vector machines (R= 0.51, R2= 0.26), and neural networks (R= 0.46, R2= 0.21) effectively predicted the daily Ambrosia pollen, where the correlation coefficient (R) and R-squared (R2) values are given in brackets. Three independent approaches-the random forests, correlation coefficients, and interaction information-were employed to rank the relative importance of the available predictors.

Combining domain filling with a self-organizing map to analyze multi-species hydrocarbon signatures on a regional scale.

For the period of the Barnett Coordinated Campaign, October 16-31, 2013, hourly concentrations for 46 volatile organic compounds (VOCs) were recorded at 14 air monitoring stations within the Barnett Shale of North Texas. These measurements are used to identify and analyze multi-species hydrocarbon signatures on a regional scale through the novel combination of two techniques: domain filling with Lagrangian trajectories and the machine learning unsupervised classification algorithm called a self-organizing map (SOM). This combination of techniques is shown to accurately identify concentration enhancements in the lightest measured alkane species at and downwind of the locations of active-permit oil and gas facilities, despite the model having no a priori knowledge of these source locations. Site comparisons further identify the SOM's ability to distinguish between signatures with differing influences from oil- and gas-related processes and from urban processes. A random forest (a machine learning supervised classification) analysis is conducted to further probe the sensitivities of the SOM classification in response to changes in any hydrocarbon species' concentration values. The random forest analysis of four representative classes finds that the SOM classification is appropriately more sensitive to changes in certain urban-related species for urban-related classes, and to changes in oil- and gas-related species for oil- and gas-related classes.

Applying Deep Neural Networks and Ensemble Machine Learning Methods to Forecast Airborne Ambrosia Pollen.

Allergies to airborne pollen are a significant issue affecting millions of Americans. Consequently, accurately predicting the daily concentration of airborne pollen is of significant public benefit in providing timely alerts. This study presents a method for the robust estimation of the concentration of airborne Ambrosia pollen using a suite of machine learning approaches including deep learning and ensemble learners. Each of these machine learning approaches utilize data from the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric weather and land surface reanalysis. The machine learning approaches used for developing a suite of empirical models are deep neural networks, extreme gradient boosting, random forests and Bayesian ridge regression methods for developing our predictive model. The training data included twenty-four years of daily pollen concentration measurements together with ECMWF weather and land surface reanalysis data from 1987 to 2011 is used to develop the machine learning predictive models. The last six years of the dataset from 2012 to 2017 is used to independently test the performance of the machine learning models. The correlation coefficients between the estimated and actual pollen abundance for the independent validation datasets for the deep neural networks, random forest, extreme gradient boosting and Bayesian ridge were 0.82, 0.81, 0.81 and 0.75 respectively, showing that machine learning can be used to effectively forecast the concentrations of airborne pollen.

Advanced Metrics for Assessing Holistic Care: The "Epidaurus 2" Project.

In response to the challenge of military traumatic brain injury and posttraumatic stress disorder, the US military developed a wide range of holistic care modalities at the new Walter Reed National Military Medical Center, Bethesda, MD, from 2001 to 2017, guided by civilian expert consultation via the Epidaurus Project. These projects spanned a range from healing buildings to wellness initiatives and healing through nature, spirituality, and the arts. The next challenge was to develop whole-body metrics to guide the use of these therapies in clinical care. Under the "Epidaurus 2" Project, a national search produced 5 advanced metrics for measuring whole-body therapeutic effects: genomics, integrated stress biomarkers, language analysis, machine learning, and "Star Glyphs." This article describes the metrics, their current use in guiding holistic care at Walter Reed, and their potential for operationalizing personalized care, patient self-management, and the improvement of public health. Development of these metrics allows the scientific integration of holistic therapies with organ-system-based care, expanding the powers of medicine.

Remote Sensing of CDOM, CDOM Spectral Slope, and Dissolved Organic Carbon in the Global Ocean.

A Global Ocean Carbon Algorithm Database (GOCAD) has been developed from over 500 oceanographic field campaigns conducted worldwide over the past 30 years including in situ reflectances and coincident satellite imagery, multi- and hyperspectral Chromophoric Dissolved Organic Matter (CDOM) absorption coefficients from 245-715 nm, CDOM spectral slopes in eight visible and ultraviolet wavebands, dissolved and particulate organic carbon (DOC and POC, respectively), and inherent optical, physical, and biogeochemical properties. From field optical and radiometric data and satellite measurements, several semi-analytical, empirical, and machine learning algorithms for retrieving global DOC, CDOM, and CDOM slope were developed, optimized for global retrieval, and validated. Global climatologies of satellite-retrieved CDOM absorption coefficient and spectral slope based on the most robust of these algorithms lag seasonal patterns of phytoplankton biomass belying Case 1 assumptions, and track terrestrial runoff on ocean basin scales. Variability in satellite retrievals of CDOM absorption and spectral slope anomalies are tightly coupled to changes in atmospheric and oceanographic conditions associated with El Niño Southern Oscillation (ENSO), strongly covary with the multivariate ENSO index in a large region of the tropical Pacific, and provide insights into the potential evolution and feedbacks related to sea surface dissolved carbon in a warming climate. Further validation of the DOC algorithm developed here is warranted to better characterize its limitations, particularly in mid-ocean gyres and the southern oceans.

Using machine learning to estimate atmospheric Ambrosia pollen concentrations in Tulsa, OK.

This article describes an example of using machine learning to estimate the abundance of airborne Ambrosia pollen for Tulsa, OK. Twenty-seven years of historical pollen observations were used. These pollen observations were combined with machine learning and a very complete meteorological and land surface context of 85 variables to estimate the daily Ambrosia abundance. The machine learning algorithms employed were Least Absolute Shrinkage and Selection Operator (LASSO), neural networks, and random forests. The best performance was obtained using random forests. The physical insights provided by the random forest are also discussed.

Phytophthora megakarya and P. palmivora, Causal Agents of Black Pod Rot, Induce Similar Plant Defense Responses Late during Infection of Susceptible Cacao Pods.

Phytophthora megakarya (Pmeg) and Phytophthora palmivora (Ppal) cause black pod rot of Theobroma cacao L. (cacao). Of these two clade 4 species, Pmeg is more virulent and is displacing Ppal in many cacao production areas in Africa. Symptoms and species specific sporangia production were compared when the two species were co-inoculated onto pod pieces in staggered 24 h time intervals. Pmeg sporangia were predominantly recovered from pod pieces with unwounded surfaces even when inoculated 24 h after Ppal. On wounded surfaces, sporangia of Ppal were predominantly recovered if the two species were simultaneously applied or Ppal was applied first but not if Pmeg was applied first. Pmeg demonstrated an advantage over Ppal when infecting un-wounded surfaces while Ppal had the advantage when infecting wounded surfaces. RNA-Seq was carried out on RNA isolated from control and Pmeg and Ppal infected pod pieces 3 days post inoculation to assess their abilities to alter/suppress cacao defense. Expression of 4,482 and 5,264 cacao genes was altered after Pmeg and Ppal infection, respectively, with most genes responding to both species. Neural network self-organizing map analyses separated the cacao RNA-Seq gene expression profiles into 24 classes, 6 of which were largely induced in response to infection. Using KEGG analysis, subsets of genes composing interrelated pathways leading to phenylpropanoid biosynthesis, ethylene and jasmonic acid biosynthesis and action, plant defense signal transduction, and endocytosis showed induction in response to infection. A large subset of genes encoding putative Pr-proteins also showed differential expression in response to infection. A subset of 36 cacao genes was used to validate the RNA-Seq expression data and compare infection induced gene expression patterns in leaves and wounded and unwounded pod husks. Expression patterns between RNA-Seq and RT-qPCR were generally reproducible. The level and timing of altered gene expression was influenced by the tissues studied and by wounding. Although, in these susceptible interactions gene expression patterns were similar, some genes did show differential expression in a Phytophthora species dependent manner. The biggest difference was the more intense changes in expression in Ppal inoculated wounded pod pieces further demonstrating its rapid progression when penetrating through wounds.

Near-Field Characterization of Methane Emission Variability from a Compressor Station Using a Model Aircraft.

A model aircraft equipped with a custom laser-based, open-path methane sensor was deployed around a natural gas compressor station to quantify the methane leak rate and its variability at a compressor station in the Barnett Shale. The open-path, laser-based sensor provides fast (10 Hz) and precise (0.1 ppmv) measurements of methane in a compact package while the remote control aircraft provides nimble and safe operation around a local source. Emission rates were measured from 22 flights over a one-week period. Mean emission rates of 14 ± 8 g CH4 s(-1) (7.4 ± 4.2 g CH4 s(-1) median) from the station were observed or approximately 0.02% of the station throughput. Significant variability in emission rates (0.3-73 g CH4 s(-1) range) was observed on time scales of hours to days, and plumes showed high spatial variability in the horizontal and vertical dimensions. Given the high spatiotemporal variability of emissions, individual measurements taken over short durations and from ground-based platforms should be used with caution when examining compressor station emissions. More generally, our results demonstrate the unique advantages and challenges of platforms like small unmanned aerial vehicles for quantifying local emission sources to the atmosphere.

Estimating the global abundance of ground level presence of particulate matter (PM2.5).

With the increasing awareness of the health impacts of particulate matter, there is a growing need to comprehend the spatial and temporal variations of the global abundance of ground level airborne particulate matter with a diameter of 2.5 microns or less (PM2.5). Here we use a suite of remote sensing and meteorological data products together with ground-based observations of particulate matter from 8,329 measurement sites in 55 countries taken 1997-2014 to train a machine-learning algorithm to estimate the daily distributions of PM2.5 from 1997 to the present. In this first paper of a series, we present the methodology and global average results from this period and demonstrate that the new PM2.5 data product can reliably represent global observations of PM2.5 for epidemiological studies.