Non-linear Machine Learning Algorithms Research Articles

Research on the Quantitative Inversion of Soil Iron Oxide Content Using Hyperspectral Remote Sensing and Machine Learning Algorithms in the Lufeng Annular Structural Area of Yunnan, China.

This study used hyperspectral remote sensing to rapidly, economically, and non-destructively determine the soil iron oxide content of the Dinosaur Valley annular tectonic region of Lufeng, Yunnan Province. The laboratory determined the iron oxide content and original spectral reflectance (OR) in 138 surface soil samples. We first subjected the OR data to Savizky-Golay smoothing, followed by four spectral transformations-continuum removal reflectance, reciprocal logarithm reflectance, standard normal variate reflectance, and first-order differential reflectance-which improved the signal-to-noise ratio of the spectral curves and highlighted the spectral features. Then, we combined the correlation coefficient method (CC), competitive adaptive reweighting algorithm, and Boruta algorithm to screen out the characteristic wavelength. From this, we constructed the linear partial least squares regression model, nonlinear random forest, and XGBoost machine learning algorithms. The results show that the CC-Boruta method can effectively remove any noise and irrelevant information to improve the model's accuracy and stability. The XGBoost nonlinear machine learning algorithm model better captures the complex nonlinear relationship between the spectra and iron oxide content, thus improving its accuracy. This provides a relevant reference for the rapid and accurate inversion of iron oxide content in soil using hyperspectral data.

Decoding wheat contamination through self-assembled whole-cell biosensor combined with linear and non-linear machine learning algorithms

The contamination of mycotoxins is a serious problem around the world. It has detrimental effects on human beings and leads to tremendous economic loss. It is essential to develop a rapid and non-destructive method for contamination recognition particularly for early alarm. In this study, the whole-cell biosensor array was constructed and employed for rapid recognition of wheat contamination by combining with machine learning algorithms. Seven key VOCs were explored through univariate coupling to multivariate analysis of orthogonal partial least squares-discrimination analysis (OPLS-DA) models. The promoters of dnaK, katG, oxyR, soxS obtained from the stress-responsive of key VOCs were fused to the bacterial operon and fabricated on the whole-cell biosensor. The constructed whole-cell biosensor array was consisted with four kinds of sensors and 18 sensor unit. The bioluminescent intensity combined with linear machine learning algorithm of partial least squares discriminant analysis (PLS-DA) and non-linear algorithms of back propagating artificial neural network (BP-ANN) and least square support vector machine (LS-SVM) were employed to establish discrimination models for mold contamination especially for early warning. The Monte-Carlo strategy was performed to generate thirty subsets for modeling to give more reliable results. As a result, the whole-cell biosensor combined with non-linear algorithm of LS-SVM was practicable for detecting mold identification for wheat early-warning with the accuracy of 97.24%. Additionally, this study provides practical and effective methods not only for wheat quality guarantee and supervision but also for other foodstuffs.

Machine learning analysis/optimization of auxetic performance of a polymeric meta-hybrid structure of re-entrant and meta-trichiral

In the ever-evolving field of materials science, the quest for innovative design solutions continues. One such area of research is the optimization of auxetic performance in meta-hybrid structures. This study delves into the application of machine learning analysis to enhance the properties of a specific meta-hybrid structure combining re-entrant and meta-trichiral designs. The polymeric meta-hybrid structure is subjected to finite element (FE) analysis to determine its Poisson's ratio. Various machine learning methods, such as linear, quadratic, and non-linear approaches, are employed to accurate forecast the Poisson's ratio of the polymeric meta-hybrid structure. Although both linear and quadratic machine learning algorithms have shown some success in predicting Poisson's ratios, they cannot be considered outstanding. As a result, a non-linear machine learning algorithm has been developed to enhance the accuracy of Poisson's ratio prediction. The results indicate that non-linear machine learning techniques out performs both linear and quadratic methods when it came to predicting Poisson's ratio. Specifically, the non-linear model achieved an impressive R-sq value of 93.68%, demonstrating its exceptional accuracy and overall excellence in fitting the data. The optimum conditions for the meta-hybrid structure are determined to be a thickness of 4.35 mm, a temperature of 15 °C, and a strain rate of 1%. Under these specific parameters, the Poisson's ratio of the meta-hybrid structure is measured to be −3.899, well outside isotropic stability limit but acceptable for auxetic structures.

Identifying Psychosocial and Ecological Determinants of Enthusiasm In Youth: Integrative Cross-Sectional Analysis Using Machine Learning.

Understanding the factors contributing to mental well-being in youth is a public health priority. Self-reported enthusiasm for the future may be a useful indicator of well-being and has been shown to forecast social and educational success. Typically, cross-domain measures of ecological and health-related factors with relevance to public policy and programming are analyzed either in isolation or in targeted models assessing bivariate interactions. Here, we capitalize on a large provincial data set and machine learning to identify the sociodemographic, experiential, behavioral, and other health-related factors most strongly associated with levels of subjective enthusiasm for the future in a large sample of elementary and secondary school students. The aim of this study was to identify the sociodemographic, experiential, behavioral, and other health-related factors associated with enthusiasm for the future in elementary and secondary school students using machine learning. We analyzed data from 13,661 participants in the 2019 Ontario Student Drug Use and Health Survey (OSDUHS) (grades 7-12) with complete data for our primary outcome: self-reported levels of enthusiasm for the future. We used 50 variables as model predictors, including demographics, perception of school experience (i.e., school connectedness and academic performance), physical activity and quantity of sleep, substance use, and physical and mental health indicators. Models were built using a nonlinear decision tree-based machine learning algorithm called extreme gradient boosting to classify students as indicating either high or low levels of enthusiasm. Shapley additive explanations (SHAP) values were used to interpret the generated models, providing a ranking of feature importance and revealing any nonlinear or interactive effects of the input variables. The top 3 contributors to higher self-rated enthusiasm for the future were higher self-rated physical health (SHAP value=0.62), feeling that one is able to discuss problems or feelings with their parents (SHAP value=0.49), and school belonging (SHAP value=0.32). Additionally, subjective social status at school was a top feature and showed nonlinear effects, with benefits to predicted enthusiasm present in the mid-to-high range of values. Using machine learning, we identified key factors related to self-reported enthusiasm for the future in a large sample of young students: perceived physical health, subjective school social status and connectedness, and quality of relationship with parents. A focus on perceptions of physical health and school connectedness should be considered central to improving the well-being of youth at the population level.

1H NMR Spectroscopy Combined with Machine-Learning Algorithm for Origin Recognition of Chinese Famous Green Tea Longjing Tea.

Premium green tea is a high-value agricultural product significantly influenced by its geographical origin, making it susceptible to food fraud. This study utilized nuclear magnetic resonance (NMR) spectroscopy to perform chemical fingerprint analysis on 78 Longjing tea (LJT) samples from both protected designation of origin (PDO) regions (Zhejiang) and non-PDO regions (Sichuan, Guangxi, and Guizhou) in China. Unsupervised algorithms and heatmaps were employed for the visual analysis of the data from PDO and non-PDO teas while exploring the feasibility of linear and nonlinear machine-learning algorithms in discriminating the origin of LJT. The findings revealed that the nonlinear model random forest (92.2%), exhibited superior performance compared to the linear model linear discriminant analysis (85.6%). The random forest model identified 15 key marker metabolites for the geographical origin of LJT, such as kaempferol glycoside, glutamine, and ECG. The results support the conclusion that the integration of NMR with machine-learning classification serves as an effective tool for the quality assessment and origin identification of LJT.

Forecasting and unveiling the impeded factors of total export of Bangladesh using nonlinear autoregressive distributed lag and machine learning algorithms

Rising global oil prices are a major challenge for an emerging oil-importing nation such as Bangladesh. The majority of prior research on the economic effects of an oil price shock has concentrated on developed countries, with emerging economies receiving comparatively less attention. Bangladesh is vulnerable to price shocks due to its rising oil consumption over the past decade. This study aims to investigate how changes in oil prices would affect Bangladesh's total export earnings and to forecast the overall export volume. This study utilized a nonlinear autoregressive distributed lag (NARDL) approach to account for the asymmetric behavior of oil prices from 1991 to 2021. To assess the accuracy of predictions, the study employed the Prophet forecasting model and the Long Short-Term Memory (LSTM) method. Additionally, the symmetry test revealed a nonlinear relationship between export volume and oil price but a linear relationship between inflation and export volume. According to the NARDL assessment, both positive and negative oil shocks increase export earnings over the long run. The short run summary clarifies that both positive and negative changes in oil prices exert a significant negative effect on exports. Also, Inflation influences export earnings negatively in the short run but positively over the long term. Moreover, using machine learning methods, it was found that the LSTM method outperforms the prophet model in prediction performance with a low root mean square error (RMSE) of 1.88. Also, the analysis revealed policymakers that the export sector requires diversification to reduce its exposure to oil price shocks.

Machine learning classification of permeable conducting spheres in air and seawater using electromagnetic pulses

Abstract This paper presents machine learning classification on simulated data of permeable conducting spheres in air and seawater irradiated by low frequency electromagnetic pulses. Classification accuracy greater than 90% was achieved. The simulated data were generated using an analytical model of a magnetic dipole in air and seawater placed 1.5–3.5 m above the center of the sphere in 50 cm increments. The spheres had radii of 40 cm and 50 cm and were of permeable materials, such as steel, and non-permeable materials, such as aluminum. A series RL circuit was analytically modeled as the transmitter coil, and an RLC circuit as the receiver coil. Additive white Gaussian noise was added to the simulated data to test the robustness of the machine learning algorithms to noise. Multiple machine learning algorithms were used for classification including a perceptron and multiclass logistic regression, which are linear models, and a neural network, 1D convolutional neural network (CNN), and 2D CNN, which are nonlinear models. Feature maps are plotted for the CNNs and provide explainability of the salient parts of the time signature and spectrogram data used for classification. The pulses investigated, which expand the literature, include a two-sided decaying exponential, Heaviside step-off, triangular, Gaussian, rectangular, modulated Gaussian, raised cosine, and rectangular down-chirp. Propagation effects, including dispersion and frequency dependent attenuation, are encapsulated by the analytical model, which was verified using finite element modeling. The results in this paper show that machine learning methods are a viable alternative to inversion of electromagnetic induction (EMI) data for metallic sphere classification, with the advantage of real-time classification without the use of a physics-based model. The nonlinear machine learning algorithms used in this work were able to accurately classify metallic spheres in seawater even with significant pulse distortion caused by dispersion and frequency dependent attenuation. This paper presents the first effort towards the use of machine learning to classify metallic objects in seawater based on EMI sensing.

Predicting Crop Yield Productivity Using Machine Learning Algorithms: A Comparison of Linear and Non-linear Approaches

Predicting crop yield productivity is crucial for farmers and the agricultural sector to gain insights into crop productivity and returns. With advancements in technology and artificial intelligence, predicting crop yield using machine learning algorithms has become an important innovation. This study aimed to predict crop yield productivity using various machine learning algorithms and techniques. The dataset was sourced from Kaggle, preprocessed, and analyzed using linear algorithms such as Linear Regression, LASSO, and Ridge, as well as non-linear algorithms including SVR, KNN Regressor, and Polynomial Regression. Mean Squared Error (MSE) was computed to evaluate algorithms performance. Comparing the efficacy of linear versus non-linear algorithms on the dataset revealed that non-linear algorithms outperformed linear ones, indicating that the dataset's non-linear nature. Therefore, non-linear machine learning algorithms like SVR, KNN Regressor, and Polynomial Regression were recommended for better accuracy. Among these, KNN Regressor performed the best with an MSE of 0.00025, followed by SVR and Polynomial Regression with MSE values of 0.00142 and 0.0024, respectively.

Neural Networks for Geospatial Data

Analysis of geospatial data has traditionally been model-based, with a mean model, customarily specified as a linear regression on the covariates, and a Gaussian process covariance model, encoding the spatial dependence. While nonlinear machine learning algorithms like neural networks are increasingly being used for spatial analysis, current approaches depart from the model-based setup and cannot explicitly incorporate spatial covariance. We propose NN-GLS, embedding neural networks directly within the traditional Gaussian process (GP) geostatistical model to accommodate nonlinear mean functions while retaining all other advantages of GP, like explicit modeling of the spatial covariance and predicting at new locations via kriging. In NN-GLS, estimation of the neural network parameters for the nonlinear mean of the Gaussian Process explicitly accounts for the spatial covariance through use of the generalized least squares (GLS) loss, thus, extending the linear case. We show that NN-GLS admits a representation as a special type of graph neural network (GNN). This connection facilitates the use of standard neural network computational techniques for irregular geospatial data, enabling novel and scalable mini-batching, backpropagation, and kriging schemes. We provide methodology to obtain uncertainty bounds for estimation and predictions from NN-GLS. Theoretically, we show that NN-GLS will be consistent for irregularly observed spatially correlated data processes. We also provide a finite sample concentration rate, which quantifies the need to accurately model the spatial covariance in neural networks for dependent data. To our knowledge, these are the first large-sample results for any neural network algorithm for irregular spatial data. We demonstrate the methodology through numerous simulations and an application to air pollution modeling. We develop a software implementation of NN-GLS in the Python package geospaNN. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

Integrating temporal decomposition and data-driven approaches for predicting coastal harmful algal blooms

Frequent coastal harmful algal blooms (HABs) threaten the ecological environment and human health. Biscayne Bay in southeastern Florida also faces algal bloom issues; however, the mechanisms driving these blooms are not fully understood, emphasizing the importance of HAB prediction for effective environmental management. The overarching goal of this study is to offer a robust HAB predictive framework and try to enhance the understanding of HAB dynamics. This study established three scenarios to predict chlorophyll-a concentrations, a recognized representative of HABs: Scenario 1 (S1) using single nonlinear machine learning (ML) algorithms, hybrid Scenario 2 (S2) combining linear models and nonlinear ML algorithms, and hybrid Scenario 3 (S3) combining temporal decomposition and ML (TD-ML) algorithms. The novel-developed S3 TD-ML hybrid models demonstrated superior predictive capabilities, achieving all R2 values above 0.9 and MAPE under 30% in tests, significantly outperforming the S1 with an average R2 of 0.16 and the S2 with an R2 of −0.06. S3 models effectively captured the algal dynamics, successfully predicting complex time series with extremes and noise. In addition, we unveiled the relationship between environmental variables and chlorophyll-a through correlation analysis and found that climate change might intensify the HABs in Biscayne Bay. This research developed a precise predictive framework for early warning and proactive management of HABs, offering potential global applicability and improved prediction accuracy to address HAB challenges.

Sex determination based on craniometric parameters: a comparative approach between linear and non-linear machine learning algorithms

Introduction: Determining sex based on cranial characteristics is of great relevance in forensic anthropology. Most studies have employed linear methods (such as logistic regression) for this estimation with accuracies around 70%, rarely exceeding 90% accuracy. Several authors have tested non-linear models such as neural networks, support vector machines, and decision trees with good results, surpassing linear models. Objective: To compare linear models (logistic regression, linear regression, and linear discriminant analysis) with non-linear models (neural networks, extreme gradient boosting, support vector machine, naive Bayes, random forest, decision tree, k-nearest neighbors, and adaptive multivariate spline regression). Materials and Methods: 241 skulls used in this study were obtained from the collection of Center for Study and Research in Anatomy and Forensic Anthropology at Tiradentes University, Farolândia campus in Aracaju, Sergipe. Each skull in the collection has secure detailed records. Eighty-nine skulls with signs of craniotomy (n=58) or damage (n=30) and one unidentified were excluded. The 152 eligible skulls underwent cranial measurements. Using the Anaconda platform and Jupyter editor, the data were divided into a training group (80% of the sample) and then were tested (20% of the sample). Eleven machine learning algorithms, including both linear and non-linear models, were applied. Results: The best machine learning algorithm was a neural network with average accuracy of 93%, after 50 runs. The difference to logistic regression, which had an accuracy of 68%, was significantly (p-value of 0.01016). Conclusion: This study demonstrated the potential of the neural network for solving the sex classification problem. The study has a limitation in that neural networks perform better with a large volume of data, and this study used data from a single center. Nevertheless, in the future, more studies should be conducted testing neural networks with larger samples and skulls from other continents.

The non-linear dynamics of South Australian regional housing markets: A machine learning approach

Traditional linear models often struggle to capture regional housing markets' complex, non-linear dynamics. This study addresses this gap by developing and applying advanced machine learning algorithms to unlock unique insights into South Australian housing price behavior. Leveraging a comprehensive dataset of over 10,000 regional house sales from 2010 to 2021, we explore the non-linear relationships between housing prices and microeconomic factors (e.g., house size, land area, building quality) and socioeconomic characteristics (e.g., proximity to amenities and income levels). Our analysis employs a multi-step approach, including feature engineering, spatial data integration, correlation tests, multilevel modeling, and non-linear machine learning algorithms including Decision Tree, Random Forest, Gradient-Boosted Tree, and Artificial Neural Network. The key finding is that machine learning models outperform traditional econometric models in predicting regional housing prices, with higher accuracy and greater goodness of fit. Furthermore, we identify specific non-linear relationships, such as the increasing marginal impact of proximity to the sea on house prices as distance decreases. These findings offer valuable insights for policymakers, real estate professionals, and stakeholders, informing regional planning, infrastructure provision, and economic development strategies. This study sheds light on the complex dynamics of South Australian housing markets and lays the foundation for further research.

Autonomous flight performance optimization of fixed-wing unmanned aerial vehicle with morphing wingtip

PurposeThe purpose of this paper is to improve autonomous flight performance of a fixed-wing unmanned aerial vehicle (UAV) via simultaneous morphing wingtip and control system design conducted with optimization, computational fluid dynamics (CFD) and machine learning approaches.Design/methodology/approachThe main wing of the UAV is redesigned with morphing wingtips capable of dihedral angle alteration by means of folding. Aircraft dynamic model is derived as equations depending only on wingtip dihedral angle via Nonlinear Least Squares regression machine learning algorithm. Data for the regression analyses are obtained by numerical (i.e. CFD) and analytical approaches. Simultaneous perturbation stochastic approximation (SPSA) is incorporated into the design process to determine the optimal wingtip dihedral angle and proportional-integral-derivative (PID) coefficients of the control system that maximizes autonomous flight performance. The performance is defined in terms of trajectory tracking quality parameters of rise time, settling time and overshoot. Obtained optimal design parameters are applied in flight simulations to test both longitudinal and lateral reference trajectory tracking.FindingsLongitudinal and lateral autonomous flight performances of the UAV are improved by redesigning the main wing with morphing wingtips and simultaneous estimation of PID coefficients and wingtip dihedral angle with SPSA optimization.Originality/valueThis paper originally discusses the simultaneous design of innovative morphing wingtip and UAV flight control system for autonomous flight performance improvement. The proposed simultaneous design idea is conducted with the SPSA optimization and a machine learning algorithm as a novel approach.

Forecasting Stock Market Realized Volatility using Random Forest and Artificial Neural Network in South Africa

Volatility is often used as a key input into several financial models, yet there is still no consensus on the best-performing model in forecasting stock market returns volatility. Conventional time series models such as GARCH are the preferred models in the literature. However, this project aims to first adopt two novel non-linear machine learning algorithms, namely the Random Forest and Artificial Neural Network (ANN). The project then compares the performance of these two models in predicting stock market realized volatility for the JSE Basic Material Index (JBIND) and the JSE Financials Index (JFIN) over a period of five years. Based on the results of the project, the Random Forest model outperformed the ANN model for both the JFIN and JBIND index. Lastly, the COVID effect on the model’s performance was also considered and the results show that the negative impact of COVID on the model’s performance is ambiguous.

Prediction of Corrosion Inhibition Efficiency Based on Machine Learning for Pyrimidine Compounds: A Comparative Study of Linear and Non-linear Algorithms

The corrosion of materials poses a significant challenge in various industries, leading to substantial economic impacts. In this context, pyrimidine compounds emerge as promising, non-toxic, cost-effective, and versatile corrosion inhibitors. However, conventional methods for identifying such inhibitors are typically time-consuming, expensive, and labor-intensive. Addressing this challenge, our study leverages machine learning (ML) to predict pyrimidine compounds corrosion inhibition efficiency (CIE). Using a quantitative structure-property relationship (QSPR) model, we compared 14 linear and 12 non-linear ML algorithms to identify the most accurate predictor of CIE. The bagging regressor model demonstrated superior performance, achieving a root mean square error (RMSE) of 5.38, a mean square error (MSE) of 28.93, a mean absolute error (MAE) of 4.23, and a mean absolute percentage error (MAPE) of 0.05 in predicting the CIE values for pyrimidine compounds. This research marks a significant advancement in corrosion science, offering a novel and efficient ML-based approach as an alternative to traditional experimental methods. It shows that machine learning can quickly and accurately determine how well organic chemical inhibitors like pyrimidine stop material corrosion. This method gives the industry a new perspective and a workable solution to a problem that has existed for a long time.
 Keywords: machine learning, corrosion inhibition, pyrimidine, QSPR model, predictive analysis

Snapshots of daily life: Situations investigated through the lens of smartphone sensing.

Daily life unfolds in a sequence of situational contexts, which are pivotal for explaining people's thoughts, feelings, and behaviors. While situational data were previously difficult to collect, the ubiquity of smartphones now opens up new opportunities for assessing situations in situ, that is, while they occur. Seizing this opportunity, the present study demonstrates how smartphones can help establish associations between the psychological perception and physical reality of situations. We employed an intensive longitudinal sampling design and investigated 9,790 situational snapshots experienced by 455 participants for 14 consecutive days. These snapshots combined self-reported situation characteristics from experience samplings with their corresponding objective cues obtained via smartphone sensing. More precisely, we extracted a total of 1,356 granular cues from different sensing modalities to account for the complexity of real-world situations. We applied linear and nonlinear machine learning algorithms to examine how well these cues predicted the perceived characteristics in terms of the Situational Eight Duty, Intellect, Adversity, Mating, pOsitivity, Negativity, Deception, Sociality (DIAMONDS), finding significant out-of-sample predictions for the five dimensions reflecting the situations' Duty, Intellect, Mating, pOsitivity, and Sociality. In a series of follow-up analyses, we further explored the data patterns captured by our models, revealing, for example, that those cues related to time and location were particularly informative of the respective situation characteristics. We conclude by interpreting the mapping between cues and characteristics in real-world situations and discussing how smartphone-based situational snapshots may push the boundaries of psychological research on situations. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Estimating ammonium changes in pilot and full-scale constructed wetlands using kinetic model, linear regression, and machine learning

Constructed wetlands (CWs) are a widely utilized nature-based wastewater treatment method for various effluents. However, their application has been more focused on pilot and full-scale CWs with substantial surface areas and extended operation times, which hold greater relevance in practical scenarios. This study used kinetics, linear regression (LR), and machine learning (ML) models to estimate effluent ammonium in pilot and full-scale CWs. From screening 1476 papers, 24 pilot and full-scale CW studies were selected to extract data containing 15 features and 975 data points. Nine models were fit to this data, revealing that linear models were less effective in capturing CW effluent compared to nonlinear ML algorithms. For training data, the Monod kinetic model predicted the poorest performance with an RMSE of 41.84 mg/L and R2 of 0.34, followed by simple LR (RMSE 24.29 mg/L and R2 0.77) and multiple LR (RMSE 22.63 mg/L and R2 0.80). In contrast, Cubist and Random Forest achieved high performances, with an average RMSE of 12.01 ± 5.38 and an average R2 of 0.93 ± 0.07 for Cubist, and an average RMSE of 15.94 ± 10.69 and an average R2 of 0.91 ± 0.08 for RF. The trained Random Forest performed the best for new data, with an R2 of 0.93 and RMSE of 13.48 mg/L. This ML-based model is a valuable tool for efficiently estimating effluent ammonium concentration in pilot and full-scale CWs, thereby facilitating the design of systems.

Effect of data preprocessing and machine learning hyperparameters on mass spectrometry imaging models

The self-organizing map (SOM) is a nonlinear machine learning algorithm that is particularly well suited for visualizing and analyzing high-dimensional, hyperspectral time-of-flight secondary ion mass spectrometry (ToF-SIMS) imaging data. Previously, we compared the capabilities of the SOM with more traditional linear techniques using ToF-SIMS imaging data. Although SOMs perform well with minimal data preprocessing and negligible hyperparameter optimization, it is important to understand how different data preprocessing methods and hyperparameter settings influence the performance of SOMs. While these investigations have been reported outside of the ToF-SIMS field, no such study has been reported for hyperspectral MSI data. To address this, we used two labeled ToF-SIMS imaging datasets, one of which was a polymer microarray dataset, while the other was semisynthetic hyperspectral data. The latter was generated using a novel algorithm that we describe here. A grid-search was used to evaluate which data preprocessing methods and SOM hyperparameters had the largest impact on the performance of the SOM. This was assessed using multiple linear regression, whereby performance metrics were regressed onto each variable defining the preprocessing-hyperparameter space. We found that preprocessing was generally more important than hyperparameter selection. We also found statistically significant interactions between several parameters studied, suggesting a complex interplay between preprocessing and hyperparameter selection. Importantly, we identified interesting trends, both dataset specific and dataset agnostic, which we describe and discuss in detail.

Modeling time series radon inventory and constraints on the submarine groundwater discharge mass balance of a well-mixed, highly dynamic estuary

Although often diffuse and over-shadowed by surface-water inputs, submarine groundwater discharge (SGD) represents a significant source of freshwater and solutes to many sensitive marine environments. Radon (222Rn) is an effective naturally occurring tracer of SGD due to its non-reactive nature and dominant terrestrial source. However, limited continuous measurements and poor constraints on temporal 222Rn variability in shallow-turbulent estuaries introduce large uncertainties to SGD rate estimation. We hypothesize that nonlinear regression analysis (e.g., generalized additive model [GAM]) and machine learning (ML) algorithms could explain radon trends and resolve limitations of the traditionally used mass balance for estimating SGD in shallow and vertically mixed estuaries, thus enabling preliminary insights related to wind-driven mixing losses using sporadic groundwater tracer measurements and publicly available hydroclimatic parameters. Cross-correlation functions and a generalized additive model [GAM] were employed to quantify atmospheric and hydrologic controls and predict, respectively, 222Rn inventory variability modulated by windspeed and direction. Two ML models, neural network (deep-learning neural network [DNN]) and decision tree (random forest [RF]) based models, were trained on an intermittent 24-month 222Rn (n = 10,660, 30 min interval) dataset (2019–2021) collected from a well-mixed northwestern Gulf of Mexico estuary (Corpus Christi Bay, Texas) with publicly available inputs (e.g., windspeed/direction, tide levels, water temperature, creek discharge and barometric pressure, n = 35,088). Both ML models can predict observed radon concentrations with r2 > 0.90, however, only the RF model is able to predict within the observed range of the 222Rn magnitude dataset and finds the relevant hydroclimatic parameters (e.g., streamflow and wind direction) highly responsible for the seasonal and daily variability in the inventory, respectively. The DNN model can more closely predict extremely high inventories, but it overpredicts lows, resulting in numerous negative values, which suggests that large data gaps in the groundwater tracer cause overfitting and underperformance of this model in a wind-mixed estuary. Using the GAM predicted 222Rn dependency on wind direction and speed, the adjusted mass balance reduces overestimation of wind-mixing losses and results in overall lower fluxes. This research provides new opportunities to: 1) predict radon inventories in regions with large temporal data gaps using only publicly available hydroclimatic parameters and 2) better constrain SGD estimates in windy coastal areas, allowing for insights that can be used to plan field excursions and management strategies for coastal water and solute budgets.

Unmanned aerial vehicle (UAV) imaging and machine learning applications for plant phenotyping

The availability of timely and accurate information about plant conditions is critical for making informed decisions for maintaining or improving agricultural productivity. However, field data collection is often laborious and time-consuming, underscoring the need for accurate and efficient methods to monitor crop growth and development. Therefore, the objective of this study was to evaluate the effectiveness of combining unmanned aerial vehicle (UAV)-based imaging and machine learning (ML) techniques for monitoring sweet corn (Zea mays var. saccharata) height, biomass, and yield. The study was conducted at the Tropical Research and Education Center (TREC) during the winter (dry) season of 2020–2021 using 16 experimental plots planted with sweet corn. The treatments were set up in a completely randomized block design (RCBD) with four irrigation treatments of 25%, 50%, 75%, and 100% full irrigation with four replications each. Field data collection included plant height, fresh and dry biomass, and yield. In addition, UAV images were collected using the DJI Matrice 210 v2 UAV (SZ DJI Technology Co., Ltd., Shenzhen, China) equipped with a MicaSense RedEdge-MX multispectral sensor (MicaSense, Seattle, WA, USA). Image processing was done with Pix4Dmapper 4.7.5 (Pix4D S.A., Prilly, Switzerland). A crop surface model, representing estimated plant height (UAVH), was calculated based on pixel-to-pixel differences between digital surface and terrain models. A simple linear regression model was used to estimate sweet corn biomass and yield from UAV images estimated plant height (UAVH). In addition, two linear algorithms known as a linear model (LM) and Lasso and elastic-net regularized generalized linear model (GLMNET) and three non-linear ML algorithms including random forest (RF), support vector machine (SVM), and k-nearest neighbor (kNN) were used to predict plant height and biomass. These algorithms were chosen due to their reliable performance and ability to learn complex non-linear relationships. Eight vegetation indices with UAVH were also used to evaluate the ML models' performance to predict plant height and biomass. Results confirmed that UAV imaging could be effectively used to estimate plant height (d = 0.99, r2 = 0.99, and MAE = 5 cm). Estimated fresh and dry biomass from UAV imaging also had good agreements with measured data with r2 values of 0.75 and 0.70, respectively. A statistically significant linear correlation between measured fresh yield at harvest and UAVH was found with d, adjusted r2, and MAE of 0.84, 0.66, and 67 g m−2, respectively. Evaluation of ML algorithms revealed that all models performed well for plant biomass estimation with d values between 0.88 and 0.99. However, the kNN and SVM outperformed all other models for biomass estimation. GMLNET performed better than other models for plant height estimation. Overall, results revealed that UAV imaging and ML models could be effectively used for monitoring plant phenotypic characteristics such as height, yield, and biomass.