Important Predictor Variables Research Articles

Integrating landmark modeling framework and machine learning algorithms for dynamic prediction of tuberculosis treatment outcomes.

This study aims to establish an informative dynamic prediction model of treatment outcomes using follow-up records of tuberculosis (TB) patients, which can timely detect cases when the current treatment plan may not be effective. We used 122267 follow-up records from 17958 new cases of pulmonary TB in the Republic of Moldova. A dynamic prediction framework integrating landmark modeling and machine learning algorithms was designed to predict patient outcomes during the course of treatment. Sensitivity and positive predictive value (PPV) were calculated to evaluate performance of the model at critical time points. New measures were defined to determine when follow-up laboratory tests should be conducted to obtain most informative results. The random-forest algorithm performed better than support vector machine and penalized multinomial logistic regression models for predicting TB treatment outcomes. For all 3 outcome classes (ie, cured, not cured, and died after 24 months following treatment initiation), sensitivity and PPV of prediction models improved as more follow-up information was collected. Specifically, sensitivity and PPV increased from 0.55 to 0.84 and from 0.32 to 0.88, respectively, for the not cured class. The dynamic prediction framework utilizes longitudinal laboratory test results to predict patient outcomes at various landmarks. Sputum culture and smear results are among the important variables for prediction; however, the most recent sputum result is not always the most informative one. This framework can potentially facilitate a more effective treatment monitoring program and provide insights for policymakers toward improved guidelines on follow-up tests.

Development of machine learning prediction models to explore nutrients predictive of cardiovascular disease using Canadian linked population-based data.

Machine learning may improve use of observational data to understand the nutritional epidemiology of cardiovascular disease (CVD) through better modelling of non-linearity, non-additivity, and dietary complexity. Our objective was to develop machine learning prediction models for exploring how nutrients are related to CVD risk and to evaluate their predictive performance. We established a population-based cohort from the Canadian Community Health Survey and measured CVD incidence and mortality from 2004 to 2018 using administrative databases of national hospital discharges and deaths. Predictors included 61 nutrition variables and fourteen socioeconomic, demographic, psychological, and behavioural variables. Conditional inference forest models were interpreted and evaluated by permutation feature importance, accumulated local effects, and predictive discrimination and calibration. A total of 12 130 individuals were included in the study. Use of supplements, caffeine, and alcohol were the most important nutrition variables for prediction of CVD. Supplement use was associated with decreased risk, caffeine was associated with increasing risk, and alcohol had a u-shaped association with risk. The model had an out-of-sample c-statistic of 0.821 (95% confidence interval = 0.801-0.842). Exploratory findings included both known and novel associations and predictive performance was competitive, suggesting that further application of machine learning to nutritional epidemiology may help elucidate risks and improve predictive models. Novelty: Machine learning prediction models were developed for CVD using dietary data. Models were interpreted with interpretable machine learning techniques, revealing diverse associations between diet and CVD. Models achieved comparable or superior predictive performance to existing CVD risk prediction models.

A pretest model of obstructive coronary artery disease based on machine learning: from the C-Strat study

Objective: To develop a pretest probability model of obstructive coronary artery disease with machine learning based on multi-site Chinese population data. Methods: Chinese regiStry in early deTection and Risk strAtificaTion of coronary plaques (C-Strat) study is a prospective multi-center cohort study, in which consecutive patients with suspected obstructive coronary artery disease and ≥64 detector row coronary computed tomography angioplasty (CCTA) evaluation were included. Data from the patients were randomly split into a training set (70%) and a test set (30%). More than 50% of coronary artery stenosis by CCTA was defined as positive outcome. A boosted ensemble algorithm (XGBoost), 10-fold cross-validation and Bayesian optimization were used to establish a new prediction model-CARDIACS(pretest probability model from Chinese registry in eARly Detection and rIsk stratificAtion of Coronary plaques Study), and a logistic regression was used to establish a model-LOGISTIC in training set. The test set was used for validation and comparison among CARDIACS, LOGISTIC, UDFM (updated Diamond-Forrester Model) and DFCASS(Diamond-Forrester and CASS). Results: The study population included 29 455 patients with age of (57.0±9.7) years and 44.8% women, of whom 19.1% (5 622/29 455) had obstructive coronary artery disease. For CARDIACS, the age, the reason for visit and the body mass index (BMI) were the most important predictive variables. In the independent test set, the area under the curve (AUC) of CARDIACS was 0.72 (95%CI 0.70-0.73), which was significantly superior to that of LOGISTIC (AUC 0.69, 95%CI 0.68-0.71, P=0.015), UDFM (AUC 0.64, 95%CI 0.62-0.65, P<0.001) and DFCASS (AUC 0.66, 95%CI 0.64-0.67, P<0.001), respectively. Conclusion: Based on Chinese population, the study developed a new pretest probability model--CARDIACS, which was superior to the traditional models. CARDIACS is expected to assist in the clinical decision-making for patients with stable chest pain.

Machine learning based rate decline prediction in unconventional reservoirs

The main objective of this paper is to develop a novel machine learning based model that can accurately predict the decline curves and EUR (Estimated Ultimate Recovery) for new wells based on the data collected from nearby wells. This is because decline curves are easier and faster alternative to complex reservoir simulators which perform computationally expensive operations. In contrast to this, decline curves require only a few parameters in the equation which can be easily collected from the existing data of the wells. In this study, first we collected the well data corresponding to well parameters such as initial monthly oil flow rate ( q i ), well completion parameters (i.e., no. of fracturing stages, completed length, amount of proppant used, amount of fracturing fluid used), well location parameters (TVD Heel-Toe Difference), reservoir fluid properties (Oil API Gravity, initial 24 h period Gas-Oil Ratio (GOR), initial 24 h period Gas Produced, initial 24 h period Oil Produced), flowing tubing pressure, casing pressure tubing size,choke size from publicly available databases of the Eagle Ford Shale formation Texas RRC (Railroad Commission of Texas). Wells were selected randomly and only those wells were finally included for the study whose data of the all the required parameters were available. The model parameters were estimated by fitting the production data to the decline curve models. Artificial Neural Network (ANN) was employed to build Machine learning models as a function of the above well parameters for the corresponding model parameters. The decline curves for new or existing wells were rapidly predicted using these models. In order to estimate the predictive accuracy of these models when applied to new or test wells cross validation technique such as k-fold cross validation was employed. These models were also used to predict EUR for the test wells. Additionally, feature selection was also done using algorithms such as Chi Square Test (χ2) and Minimum Redundancy Maximum Relevance (MRMR) Algorithm to determine the relative importance of predictor variables in predicting EUR. The predictor variables were successfully linked to SEDM (Stretched Exponential Decline Model) decline curve parameters (n and τ) in a random set of oil field well data. The relative influences of various well parameters were also examined to determine the hidden relationship between them. The novelty in this study lies in the algorithm and dataset that we used for the rate decline prediction in Eagle Ford data set. Although, this paper has referenced some previous papers where machine learning has been used to make prediction, but this paper presents use of new algorithm as well as a new dataset. As more data gets available, there is definitely extra room for further data analysis and improved results using machine learning methods. In future, the number of wells can be increased and the updated results can again be submitted after investing more time. It should be noted here that data downloading and preparing takes most of the time for such study especially when dealing with oil and gas data.

Predicting medical usage rate at mass gathering events in Belgium: development and validation of a nonlinear multivariable regression model

BackgroundEvery year, volunteers of the Belgian Red Cross provide onsite medical care at more than 8000 mass gathering events and other manifestations. Today standardized planning tools for optimal preventive medical resource use during these events are lacking. This study aimed to develop and validate a prediction model of patient presentation rate (PPR) and transfer to hospital rate (TTHR) at mass gatherings in Belgium.MethodsMore than 200,000 medical interventions from 2006 to 2018 were pooled in a database. We used a subset of 28 different mass gatherings (194 unique events) to develop a nonlinear prediction model. Using regression trees, we identified potential predictors for PPR and TTHR at these mass gatherings. The additional effect of ambient temperature was studied by linear regression analysis. Finally, we validated the prediction models using two other subsets of the database.ResultsThe regression tree for PPR consisted of 7 splits, with mass gathering category as the most important predictor variable. Other predictor variables were attendance, number of days, and age class. Ambient temperature was positively associated with PPR at outdoor events in summer. Calibration of the model revealed an R2 of 0.68 (95% confidence interval 0.60–0.75). For TTHR, the most determining predictor variables were mass gathering category and predicted PPR (R2 = 0.48). External validation indicated limited predictive value for other events (R2 = 0.02 for PPR; R2 = 0.03 for TTHR).ConclusionsOur nonlinear model performed well in predicting PPR at the events used to build the model on, but had poor predictive value for other mass gatherings. The mass gathering categories “outdoor music” and “sports event” warrant further splitting in subcategories, and variables such as attendance, temperature and resource deployment need to be better recorded in the future to optimize prediction of medical usage rates, and hence, of resources needed for onsite emergency medical care.

Population median imputation was noninferior to complex approaches for imputing missing values in cardiovascular prediction models in clinical practice

ObjectivesTo compare the validity and robustness of five methods for handling missing characteristics when using cardiovascular disease risk prediction models for individual patients in a real-world clinical setting. Study design and settingThe performance of the missing data methods was assessed using data from the Swedish National Diabetes Registry (n = 419,533) with external validation using the Scottish Care Information ˗ diabetes database (n = 226,953). Five methods for handling missing data were compared. Two methods using submodels for each combination of available data, two imputation methods: conditional imputation and median imputation, and one alternative modeling method, called the naïve approach, based on hazard ratios and populations statistics of known risk factors only. The validity was compared using calibration plots and c-statistics. ResultsC-statistics were similar across methods in both development and validation data sets, that is, 0.82 (95% CI 0.82–0.83) in the Swedish National Diabetes Registry and 0.74 (95% CI 0.74–0.75) in Scottish Care Information-diabetes database. Differences were only observed after random introduction of missing data in the most important predictor variable (i.e., age). ConclusionValidity and robustness of median imputation was not dissimilar to more complex methods for handling missing values, provided that the most important predictor variables, such as age, are not missing.

Informative predictors of pregnancy after first IVF cycle using eIVF practice highway electronic health records

The aim of this study is to determine the most informative pre- and in-cycle variables for predicting success for a first autologous oocyte in-vitro fertilization (IVF) cycle. This is a retrospective study using 22,413 first autologous oocyte IVF cycles from 2001 to 2018. Models were developed to predict pregnancy following an IVF cycle with a fresh embryo transfer. The importance of each variable was determined by its coefficient in a logistic regression model and the prediction accuracy based on different variable sets was reported. The area under the receiver operating characteristic curve (AUC) on a validation patient cohort was the metric for prediction accuracy. Three factors were found to be of importance when predicting IVF success: age in three groups (38–40, 41–42, and above 42 years old), number of transferred embryos, and number of cryopreserved embryos. For predicting first-cycle IVF pregnancy using all available variables, the predictive model achieved an AUC of 68% + /− 0.01%. A parsimonious predictive model utilizing age (38–40, 41–42, and above 42 years old), number of transferred embryos, and number of cryopreserved embryos achieved an AUC of 65% + /− 0.01%. The proposed models accurately predict a single IVF cycle pregnancy outcome and identify important predictive variables associated with the outcome. These models are limited to predicting pregnancy immediately after the IVF cycle and not live birth. These models do not include indicators of multiple gestation and are not intended for clinical application.

Vegetation Mapping in the Permafrost Region: A Case Study on the Central Qinghai-Tibet Plateau

An accurate and detailed vegetation map is of crucial significance for understanding the spatial heterogeneity of subsurfaces, which can help to characterize the thermal state of permafrost. The absence of an alpine swamp meadow (ASM) type, or an insufficient resolution (usually km-level) to capture the spatial distribution of the ASM, greatly limits the availability of existing vegetation maps in permafrost modeling of the Qinghai-Tibet Plateau (QTP). This study generated a map of the vegetation type at a spatial resolution of 30 m on the central QTP. The random forest (RF) classification approach was employed to map the vegetation based on 319 ground-truth samples, combined with a set of input variables derived from the visible, infrared, and thermal Landsat-8 images. Validation using a train-test split (i.e., 70% of the samples were randomly selected to train the RF model, while the remaining 30% were used for validation and a total of 1000 runs) showed that the average overall accuracy and Kappa coefficient of the RF approach were 0.78 (0.68–0.85) and 0.69 (0.64–0.74), respectively. The confusion matrix showed that the overall accuracy and Kappa coefficient of the predicted vegetation map reached 0.848 (0.844–0.852) and 0.790 (0.785–0.796), respectively. The user accuracies for the ASM, alpine meadow, alpine steppe, and alpine desert were 95.0%, 83.3%, 82.4%, and 86.7%, respectively. The most important variables for vegetation type prediction were two vegetation indices, i.e., NDVI and EVI. The surface reflectance of visible and shortwave infrared bands showed a secondary contribution, and the brightness temperature and the surface temperature of the thermal infrared bands showed little contribution. The dominant vegetation in the study area is alpine steppe and alpine desert. The results of this study can provide an accurate and detailed vegetation map, especially for the distribution of the ASM, which can help to improve further permafrost studies.

Prediction of pressure gradient for oil-water flow: A comprehensive analysis on the performance of machine learning algorithms

Pressure gradient (PG) in liquid-liquid flow is one of the key components to design an energy-efficient transportation system for wellbores. This study aims to develop five robust machine learning (ML) algorithms and their fusions for a wide range of flow patterns (FP) regimes. The MLs include Support Vector Machine (SVM), Gaussian Process (GP), Random Forest (RF), Artificial Neural Network (ANN), k-Nearest Neighbor (kNN), and fusions of these five MLs. A total of eleven hundred experimental data points for nine FPs (two stratified and seven dispersed patterns) in horizontal wellbores are used to develop the MLs. The MLs' performance is evaluated using the metrics including mean absolute percentage error (MAPE), median absolute percentage error (MdAPE), coefficient of variation of root mean squared error (CV-RMSE), and adjusted coefficient of determination. The evaluation metrics are cross-validated using a repeated train-test split strategy. Seven important predictor variables are identified using a supervised feature selection approach: oil and water velocities, FP, input diameter, oil and water density, and oil viscosity. The results show that the high PG prediction accuracy can be achieved using GP compared to other MLs except for the ML-fusions (p < 0.05). A Friedman's test and Wilcoxon Sign-Rank post hoc analysis with Bonferroni correction show that PG prediction errors using GP are significantly lower than using the ANN model (p < 0.05). The values are 18.44 % and 23.9 % for CV-RMSE, 11.6 % and 10.06 % for MAPE, and 7.5 % and 6.75 % for MdAPE, using ANN and GP, respectively. While the previous studies mostly used ANN to demonstrate the capability of MLs to predict PG over the mechanistic or correlation-based models, the present research has shown that GP is even better than ANN using a wide range of FPs and a large data set. • Machine-learning algorithms are developed to predict the pressure gradient. • The Gaussian Process model produces significantly lower prediction error. • The median absolute percentage error using Gaussian Process is 6.75 %. • More than 95 % variation of pressure gradient can be explained using seven variables.

Association between seroprevalence and risk factors for brucellosis in small ruminants of Gujarat

A study was aimed to estimate the seroprevalence and assess the potential risk factors of brucellosis among sheep and goats in Gujarat from 2013 to 2018. Atotal of 2445 sera (goats=1373 and sheep=1072) were subjected to Rose Bengal plate test (RBPT) and indirect enzyme-linked immunosorbent assay (iELISA). A total of seven important risk factors or predictor variables were identified and variables with p<0.10 were retained for univariate logistic regression analysis. The apparent and true seroprevalence were 6.70 and 7.84% in goats and 13.24 and 15.50% in sheep, respectively. In goats, univariate logistic regression analysis affirmed that only two factors i.e., history of disease (odds ratio, OR- 5.03) and larger flock size (OR-3.03) had significant (p<0.001) association with the seroprevalence of brucellosis. In addition to these two, location (OR- 2.71) was also significantly associated (P<0.001) with seroprevalence of brucellosis in sheep. Least Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values (625.61 and 677.86) as indicators of best correlation were obtained when breed, sex, history of disease and flock size were used as predictor variables for goats. In sheep, least AIC and BIC values (688.55 and 728.36) were obtained with the model incorporating breed, history of disease, region of sampling and flock size as predictors. To conclude, in goats and sheep, three common factors viz., breed, history and flock size were significantly associated with seroprevalence of brucellosis.

Determining the distributions of plant communities in subantarctic vegetation using species distribution models

Context Mapping of vegetation is important in understanding its dynamics in relation to climate change and disturbance. We investigated using species distribution models to predict plant species assemblages in a subantarctic environment where traditional image interpretation methods of vegetation mapping are limited by image availability and ability to discriminate vegetation types. Aims We test the efficacy for mapping of modelling the range and core range of common species. We also determine the relative importance of predictor variables for each of nine species. Methods We used random forest models to predict the total range and core range (&gt;25% projected foliage cover) of nine potentially dominant plant species and determined the contributions of predictor variables to the models for each species. Key results Widespread species with extensively overlapping ranges were spatially more partitioned with modelling based on core range than with presence or absence modelling. The core range input produced a vegetation map that better approximated observed vegetation patterns than that from presence or absence data. The most important predictor variable varied between species, with elevation, distance from coast, latitude and an across island gradient (similar to longitude) being most influential. Conclusions Species distribution models using three categories (absent, &lt;25% cover, ≥25% cover) and topographic variables derived from a digital elevation model can be used to model the distribution of vegetation assemblages in situations where presence or absence species models cannot discriminate assemblages. Implications Readily collected point location species data could be used to investigate change over time in the spatial extent of both species and vegetation types.

Development and validation of specific post-transplant risk scores according to the circulatory support status at transplant: A UNOS cohort analysis

The clinical use of post-transplant risk scores is limited by their poor statistical performance. We hypothesized that developing specific prognostic models for each type of circulatory support at transplant may improve risk stratification. We analyzed the UNOS database including contemporary, first, non-combined heart transplantations (2013–2018). The primary endpoint was death or retransplantation during the first year post-transplant. Three different circulatory support statuses at transplant were considered: no support, durable mechanical support and temporary support (inotropes, temporary mechanical support). We generated 1000 bootstrap samples that we randomly split into derivation and test sets. In each sample, we derived an overall model and three specific models (one for each type of circulatory support) using Cox regressions, and compared, in the test set, their statistical performance for each type of circulatory support. A total of 13,729 patients were included; 1220 patients (8.9%) met the composite endpoint. Circulatory support status at transplant was associated with important differences in baseline characteristics and distinct prognosis ( P = 0.01), interacted significantly with important predictive variables included in the overall model, and had a major impact on post-transplant predictive models (type of variables included and their corresponding hazard ratios). However, specific models significantly improved risk stratification (discrimination, reclassification indices, calibration) compared to overall models in a very limited proportion of bootstrap samples (< 15%). These results were consistent across several sensitivity analyzes ( Fig. 1 ). Circulatory support status at transplant reflected different disease states that influenced predictive models. However, developing specific models for each circulatory support status did not significantly improve risk stratification.

Associating multiple mycotoxin exposure and health outcomes: current statistical approaches and challenges

Mycotoxin contamination is a global challenge to food safety and population health. A diversity of adverse effects in human health such as organ damage, immunity disorders and carcinogenesis are attributed to acute and chronic exposure to mycotoxins. While there is a high likelihood of mycotoxin co-occurrence in the daily diet, multiple mycotoxin exposures represent a considerable challenge in understanding the accumulative effects of groups of exposures on health outcomes. Nevertheless, previous studies on mycotoxin exposure-health outcome associations have focused on a single or a limited number of exposures. To guide multi-exposure assessment, careful considerations of statistical approaches available are required. In addition, the issue of multicollinearity in high-dimensional settings of multiple exposure analysis underlies the controversy surrounding the reliability and consistency of statistical conclusions about the exposure-health outcome associations. Conventional approaches such as generalised linear regressions (GLR) in conjunction with regularisation methods, including ridge regression, lasso and elastic net, offer some clear advantages in terms of results’ interpretation and model selection. However, when highly-correlated variables are observed, these methods have shown a low specificity in variable selection. Principal component analysis (PCA) that has been widely used as a dimensionality reduction technique also has the limitation to identify important predictor variables as this approach may overlook the associations between certain components and health outcomes. Recently, some alternative approaches have been introduced to address the issues of high dimensionality and highly-correlated data in the context of epidemiological and environmental research. Two of the noticeable approaches are weighted quantile sum regression (WQSR) and Bayesian kernel machine regression (BKMR). Combining different methods of inference allows us to interpret the role of certain exposures, their interactions and the combined effects on human health under diverse statistical perspectives, which ultimately facilitate the construction of the toxicological profile of multiple mycotoxins’ exposure.

Do Patient Personality Traits and Self-Reported Physical and Psychosocial Symptoms Help to Predict Hallux Valgus Surgery Outcomes?

As many as 10% of patients remain unsatisfied after hallux valgus surgery. We explored the effects of patient personality traits and other preoperative patient characteristics on patient-reported outcomes following surgery. Eighty consecutive adult patients (mean age 45 ± 14 years, 91% female [73/80]) undergoing scarf bunionectomy at our practice were prospectively enrolled from January 2016 to January 2017 and followed for 12 months. Predictor variables included preoperative physical and psychosocial complaints (determined via Brief Battery for Health Improvement-2 questionnaire), patient aggression level, and personality traits (extraversion, agreeableness, conscientiousness, emotional stability and openness). Primary outcome measures included the Foot and Ankle Outcome Score (FAOS) with its 5 subscales, and patient satisfaction. Multiple multivariable regression models were used to determine preoperative patient characteristics associated with FAOS outcome and satisfaction at 12 months. Seventy subjects (70/80, 87.5%) completed the study. All patients experienced technically successful surgery. In the multivariable regression analyses, none of the combinations of potentially important predictor variables explained more than 19.8% of the variance in any of the 5 FAOS subscales at 12 months (range: 6.1%-19.8%). Furthermore, no predictor was associated with patient satisfaction in either the univariate or multivariable analyses. We conclude that patient personality traits, aggression level, and self-reported physical and psychological symptoms do very little to predict outcomes in hallux valgus surgery. As healthcare delivery in the United States has increasingly prioritized patient satisfaction, we will need to broaden the quest for predictors associated with our best (and worst) patient-reported outcomes after hallux valgus surgery.

Modelling and Mapping Total and Bioaccessible Arsenic and Lead in Stoke-on-Trent and Their Relationships with Industry

This study was based on a geochemical soil survey of Stoke-on-Trent in the UK of 747 surface soil samples analysed for 53 elements. A subset of 50 of these soil samples were analysed for their bioaccessible As and Pb content using the Unified Barge Method. Random Forest modelling, using the total element data as predictor variables, was used to predict bioaccessible As and Pb for all 747 samples. Random Forest modelling, using inverse distance weighed predictors and bedrock and superficial geology, was also used to map both total and bioaccessible As and Pb on a 400 × 400 spatial prediction grid with a 50 m resolution. The predicted bioaccessible As ranged from ca. 1 to 8 mg/kg and the total As ca. 8 to 45 mg/kg. The bioaccessible Pb and the total Pb both covered the range ca. 16–1200 mg/kg, with the highest values for both forms of Pb showing similar spatial distributions. Predictor variable importance and information on past industry suggest that the source of both of these elements is driven by anthropogenic causes.

Serological Survey of Canine Vector-Borne Infections in North-Center Spain.

Various factors are currently causing an increase in vector-borne parasitic diseases at a global scale; among them, some stand out, such as climatic disturbances derived from global change, the increase in movements of reservoir animals, or changes in land made by human activity. In the European continent, there have been an increasing number of epidemiological studies focused on the detection of these diseases, especially in dogs. In Spain, there are few epidemiological studies focused on the evaluation of the biotic and abiotic factors that may influence the distribution, such as climatic zones, orography, or presence of water reservoirs. The aim of this study was to analyze the prevalence and distribution of several canine vector-borne diseases caused by Dirofilaria immitis, Leishmania infantum, Anaplasma platys, and Ehrlichia canis in the autonomous community of Castilla y León, the largest region of the Iberian Peninsula, providing a geospatial approach based on a geographic information system (GIS) analysis. Blood from a total of 1,475 domestic dogs from the nine provinces of Castilla y León were analyzed. Also, a GIS analysis of the sample locations was carried out, taking into account the most important predictor variables. The prevalence in dogs infected by D. immitis was 7.19%, and the seroprevalence by L. infantum was 4.61 and 1.56% for A. platys and E. canis. Most of the infected animals were located in areas with stagnant water, irrigated agriculture, or riverbanks, always close to forest and woodland vegetation. These results indicate that dogs living in Castilla y León should take prophylactic measures to avoid infections.

Prediction of soil organic matter content by combining data from Nix ProTM color sensor and portable X-ray fluorescence spectrometry in tropical soils

Soil organic matter (SOM) measurement is of great agricultural and environmental importance. Thus, the development of rapid, environmentally-friendly, economical and reliable assessment methods is challenging. Soil proximal sensors have become an important approach for SOM prediction worldwide, but require regional calibration. This work aimed to assess the efficiency of SOM content prediction using the Nix Pro™ color sensor and portable X-ray fluorescence (pXRF) spectrometry, either separately or combined. The type of soil horizon collected (A or B) was used as auxiliary input data. A total of 705 Brazilian variable soil samples were analyzed in the laboratory for SOM content and scanned by Nix Pro™ and pXRF. Via Nix Pro™, samples were analyzed both dry and moist since moisture changes their color. Prediction models were built using 70% of the data via the stepwise multiple linear regression (SMLR), support vector machine with linear kernel (SVM) and random forest (RF). Validation was performed with the remaining 30% of the data through the coefficient of determination (R2), the root mean square error (RMSE) and the residual prediction deviation (RPD). SOM content was predicted with good accuracy (R2 = 0.73, RMSE = 1.09% and RPD = 2.00) using the RF algorithm trained with combined data from the Nix Pro™ and pXRF sensors. Soil horizons and Ca content were the two most important predictor variables. The combination of data obtained by Nix Pro™ and pXRF yielded accurate SOM predictions for a wide variety of Brazilian soils, in addition to being environmentally-friendly, without generating chemical waste.

Machine Learning Models of COVID-19 Cases in the United States: A Study of Initial Lockdown and Reopen Regimes

The purpose of this paper is to model the cases of COVID-19 in the United States from 13 March 2020 to 31 May 2020. Our novel contribution is that we have obtained highly accurate models focused on two different regimes, lockdown and reopen, modeling each regime separately. The predictor variables include aggregated individual movement as well as state population density, health rank, climate temperature, and political color. We apply a variety of machine learning methods to each regime: Multiple Regression, Ridge Regression, Elastic Net Regression, Generalized Additive Model, Gradient Boosted Machine, Regression Tree, Neural Network, and Random Forest. We discover that Gradient Boosted Machines are the most accurate in both regimes. The best models achieve a variance explained of 95.2% in the lockdown regime and 99.2% in the reopen regime. We describe the influence of the predictor variables as they change from regime to regime. Notably, we identify individual person movement, as tracked by GPS data, to be an important predictor variable. We conclude that government lockdowns are an extremely important de-densification strategy. Implications and questions for future research are discussed.

Impacts of Soil Properties, Topography, and Environmental Features on Soil Water Holding Capacities (SWHCs) and Their Interrelationships

Soil water holding capacities (SWHCs) are among the most important factors for understanding the water cycle in forested catchments because they control available plant water that supports evapotranspiration. The direct determination of SWHCs, however, is time consuming and expensive, so many pedotransfer functions (PTFs) and digital soil mapping (DSM) models have been developed for predicting SWHCs. Thus, it is important to select the correct soil properties, topographies, and environmental features when developing a prediction model, as well as to understand the interrelationships among variables. In this study, we collected soil samples at 971 forest sites and developed PTF and DSM models for predicting three kinds of SWHCs: saturated water content (θS) and water content at pF1.8 and pF2.7 (θ1.8 and θ2.7). Important explanatory variables for SWHC prediction were selected from two variable importance analyses. Correlation matrix and sensitivity analysis based on the developed models showed that, as the matric suction changed, the soil physical and chemical properties that influence the SWHCs changed, i.e., soil structure rather than soil particle distribution at θS, coarse soil particles at θ1.8, and finer soil particle at θ2.7. In addition, organic matter had a considerable influence on all SWHCs. Among the topographic features, elevation was the most influential, and it was closely related to the geological variability of bedrock and soil properties. Aspect was highly related to vegetation, confirming that it was an important variable for DSM modeling. Information about important variables and their interrelationship can be used to strengthen PTFs and DSM models for future research.

Prediction of butt rot volume in Norway spruce forest stands using harvester, remotely sensed and environmental data

Butt rot (BR) damage of a tree results from a decay caused by a pathogenic fungus. BR damages associated with Norway spruce (Picea abies [L.] Karst.) account for considerable economic losses in timber production across the northern hemisphere. While information on BR damages is critical for optimal decision-making in forest management, maps of BR damages are typically lacking in forest information systems. Timber volume damaged by BR was predicted at the stand-level in Norway using harvester information of 186,026 stems (clear-cuts), remotely sensed, and environmental data (e.g. climate and terrain characteristics). This study utilized Random Forests models with two sets of predictor variables: (1) predictor variables available after harvest (theoretical case) and (2) predictor variables available prior to harvest (mapping case). Our findings showed that forest attributes characterizing the maturity of forest, such as remote sensing-based height, harvested timber volume and quadratic mean diameter at breast height, were among the most important predictor variables. Remotely sensed predictor variables obtained from airborne laser scanning data and Sentinel-2 imagery were more important than the environmental variables. The theoretical case with a leave-stand-out cross-validation resulted in an RMSE of 11.4 m3 · ha−1 (pseudo-R2: 0.66) whereas the mapping case resulted in a pseudo-R2 of 0.60. When spatially distinct clusters of harvested forest stands were used as units in the cross-validation, the RMSE value and pseudo-R2 associated with the mapping case were 15.6 m3 · ha−1 and 0.37, respectively. The findings associated with the different cross-validation schemes indicated that the knowledge about the BR status of spatially close stands is of high importance for obtaining satisfactory error rates in the mapping of BR damages.