Support Vector Machine Regression Model Research Articles

Prediction of real contact area of fractal rough surface based on FEM contact model results and support vector machine algorithm

The metal-metal contact nonlinearity can degrade the performance of communication system, which has become a hot issue in the electrical contact field. The key problem is to establish the relationship between the parameters of rough surface and the real contact area under the preload. The surface roughness parameters are dependent on the metal material and processing technology etc. The real contact area is generally calculated by the finite element method (FEM). However, there exist some difficulties for FEM such as exact model of rough surface and tedious calculation in the engineering. Aimed at this problem, we propose a new method to predict the real contact area of fractal rough surface, and establish the least-square support vector machines (SVM) regression model and optimize the hyper-parameters by the combination of the coupled simulated annealing and grid searching algorithms. The model can predict the real contact area of metal-metal contact for different working conditions, and address the problem to establish the relationship between the parameters of rough surfaces and the real contact area under the preload. The work lays good foundation for analysis of metal-metal contact nonlinearity.

Analysis and Optimization of Groundwater Distribution Using SVM and Neural Networks

Water is the elixir of life. Gradual increase in demand for water, whether for agricultural, personal or industrial use has led to the depletion of clean water resources. Hence, to battle this water crisis, we propose to build a model using machine learning. The goal of the model is to predict the groundwater levels of different areas in Delhi with respect to various parameters that affect the groundwater levels in the environment. To find the best machine learning technique to perform the prediction of groundwater, we initially perform a comparative analysis of four machine learning techniques: Support vector machine, Artificial neural networks, Random forests and Linear regression models. We proceed with the models that give satisfactory results. We perform forecasting using the ARIMA model to predict the parameters that affect groundwater levels across the next ten years. The forecasting of parameters resulted in increasing the database by two times, hence we performed the prediction of the water table using the models Support vector machine, Artificial neural networks and Linear regression models. Thus, concluded that the Artificial neural networks model performs better than the Support vector machine and Linear regression models. It presented with a mean absolute error of 85.26 as opposed to linear regression model that presented with a mean absolute error of 94.95 and Support vector machine with a mean absolute error of 127.74. Further these predictions are used to build optimized groundwater distribution routes using the optimization algorithm vehicle routing problem. If these routes are implemented using proper infrastructure, we can provide the resource of groundwater to the areas that are experiencing scarcity of water.

Estimation of the Microbiological Quality of Meat Using Rapid and Non-Invasive Spectroscopic Sensors

Spectroscopic methods in tandem with machine learning methodologies have attracted considerable research interest for the estimation of food quality. The objective of this study was the evaluation of Fourier transform infrared (FTIR) spectroscopy and multispectral imaging (MSI) coupled with appropriate machine learning regression algorithms for assessing meat microbiological quality. For this purpose, minced pork patties were stored aerobically and under modified atmosphere packaging (MAP) conditions, at isothermal and dynamic temperature conditions. At regular time intervals during storage, samples were subjected to (i) microbiological analysis, (ii) FTIR measurements and (iii) MSI acquisition. The collected FTIR data were processed by feature extraction methods to reduce dimensionality, and subsequently Support Vector Machines (SVM) regression models were trained using spectral features (FTIR and MSI) to estimate microbiological quality of meat (microbial population). The regression models were evaluated with different experimental replicates using distinct meat batches. The performance of the models was evaluated in terms of correlation coefficient (r), root mean square error (RMSE), mean absolute error (MAE) and residual prediction deviation (RPD). The RMSE values for the microbial population estimation models using FTIR were 1.268 and 1.024 for aerobic and MAP storage, respectively. The performance in terms of RMSE for the MSI-based models was 1.144 for aerobic and 0.923 for MAP storage, while the combination of FTIR and MSI spectra resulted in models with RMSE equal to 1.146 for aerobic and 0.886 for MAP storage. The experimental results demonstrated the potential of estimating the microbiological quality of minced pork meat from spectroscopic data.

Forecasting of Commodity Future Index using a Hybrid Regression Model based on Support Vector Machine and Grey Wolf Optimization Algorithm

The forecasting and investigation of finance time series data are hard, and are the most confounded works pertained with investor decision. In this paper, an economic derivative instrument for Multi Commodity Exchange (MCX) index of CRUDEOIL is estimated by utilizing forecasting models based on recently formulated artificial intelligence (AI) approaches. These approaches have been appeared to perform astoundingly well in different optimization problems. Specifically, a novel hybrid forecasting model is designed by combining the support vector machine (SVM) and grey wolf optimization (GWO) and it is named as hybrid SVM-GWO. The presented hybrid SVM-GWO model eliminates the user determined control parameter, which is needed for other AI techniques. The practicality and proficiency of the presented SVM-GWO regression method is evaluated by predicting the everyday close price of CRUDEOIL index traded in the MCX of India Limited. The exploratory outcomes depicts that the present hybrid SVM-GWO technique is viable and outperforms superior to the conventional SVM, hybrid SVM-TLBO and SVM-PSO regression models.

Performance Evaluation of Diagnosis Chronic Kidney Disease using Support Vector Machine and Logistic Regression Model

Performance Evaluation of Diagnosis Chronic Kidney Disease using Support Vector Machine and Logistic Regression Model

Nondestructive Determination of Nitrogen, Phosphorus and Potassium Contents in Greenhouse Tomato Plants Based on Multispectral Three-Dimensional Imaging.

Measurement of plant nitrogen (N), phosphorus (P), and potassium (K) levels are important for determining precise fertilization management approaches for crops cultivated in greenhouses. To accurately, rapidly, stably, and nondestructively measure the NPK levels in tomato plants, a nondestructive determination method based on multispectral three-dimensional (3D) imaging was proposed. Multiview RGB-D images and multispectral images were synchronously collected, and the plant multispectral reflectance was registered to the depth coordinates according to Fourier transform principles. Based on the Kinect sensor pose estimation and self-calibration, the unified transformation of the multiview point cloud coordinate system was realized. Finally, the iterative closest point (ICP) algorithm was used for the precise registration of multiview point clouds and the reconstruction of plant multispectral 3D point cloud models. Using the normalized grayscale similarity coefficient, the degree of spectral overlap, and the Hausdorff distance set, the accuracy of the reconstructed multispectral 3D point clouds was quantitatively evaluated, the average value was 0.9116, 0.9343 and 0.41 cm, respectively. The results indicated that the multispectral reflectance could be registered to the Kinect depth coordinates accurately based on the Fourier transform principles, the reconstruction accuracy of the multispectral 3D point cloud model met the model reconstruction needs of tomato plants. Using back-propagation artificial neural network (BPANN), support vector machine regression (SVMR), and gaussian process regression (GPR) methods, determination models for the NPK contents in tomato plants based on the reflectance characteristics of plant multispectral 3D point cloud models were separately constructed. The relative error (RE) of the N content by BPANN, SVMR and GPR prediction models were 2.27%, 7.46% and 4.03%, respectively. The RE of the P content by BPANN, SVMR and GPR prediction models were 3.32%, 8.92% and 8.41%, respectively. The RE of the K content by BPANN, SVMR and GPR prediction models were 3.27%, 5.73% and 3.32%, respectively. These models provided highly efficient and accurate measurements of the NPK contents in tomato plants. The NPK contents determination performance of these models were more stable than those of single-view models.

A machine learning approach to filtrate loss determination and test automation for drilling and completion fluids

Drilling fluid property characterization currently involves several manually executed analytical tests, conducted in accordance with American Petroleum Institute (API) recommended practices 13B-1 and 13B-2. Standard (API) and high-pressure, high-temperature (HPHT) filter press units are used for filtrate loss (FL) measurements. However, these test methods have certain important disadvantages. FL tests are conducted at standardized conditions that generally do not reflect downhole oil and gas well conditions in terms of downhole pressure, temperature and filter medium encountered, present safety concerns due to their elevated pressure and temperature, and are performed only infrequently by a human mud engineer. The human measurement also introduces concerns around inaccuracies due to inconsistent practices and interpretation bias of the results. In addition, FL measurements are hard to automate given their manual, human-centric operating tasks.In this paper, we investigate if it is really necessary to automate FL measurements, or if the (strictly qualitative) information they provide can be obtained in a smarter, more advanced way. For this purpose, the relationship between fluid properties was investigated in detail using machine learning and deep learning techniques. Random forest (RF), XGBoost (XGB), support vector machine (SVM), multilayer perceptron (MLP) and multi-linear regression models were trained and tested to predict API and HPHT FL of water-based muds (WBM) based on fluid property readings of rheology, density, and temperature. A similar approach was also used for HPHT filtrate loss prediction of oil-based muds (OBM), taking into account their electrical stability and water content. A key advantage of this approach is that these WBM and OBM fluid properties can be obtained in real-time with measurements that are relatively simple and easy to automate (e.g. obtaining fluid density automatically and continuously from a Coriolis mass-flow meter measurement). Thus, real-time assessment of API and HPHT FL becomes possible without ever having to actually carry out any filter press measurements, thereby also eliminating the need to directly automate these measurements. The models were verified by dedicated laboratory experiments. The developed models estimated API and HPHT FL of WBM, and HPHT FL of OBM with mean absolute errors (MAE) of 0.56 ml/30min, 1.15 ml/30min and 0.79 ml/30min respectively, well within the measurement accuracy of the observations by a human mud engineer.

Eco-Friendly Estimation of Heavy Metal Contents in Grapevine Foliage Using In-Field Hyperspectral Data and Multivariate Analysis.

Heavy metal monitoring in food-producing ecosystems can play an important role in human health safety. Since they are able to interfere with plants’ physiochemical characteristics, which influence the optical properties of leaves, they can be measured by in-field spectroscopy. In this study, the predictive power of spectroscopic data is examined. Five treatments of heavy metal stress (Cu, Zn, Pb, Cr, and Cd) were applied to grapevine seedlings and hyperspectral data (350–2500 nm), and heavy metal contents were collected based on in-field and laboratory experiments. The partial least squares (PLS) method was used as a feature selection technique, and multiple linear regressions (MLR) and support vector machine (SVM) regression methods were applied for modelling purposes. Based on the PLS results, the wavelengths in the vicinity of 2431, 809, 489, and 616 nm; 2032, 883, 665, 564, 688, and 437 nm; 1865, 728, 692, 683, and 356 nm; 863, 2044, 415, 652, 713, and 1036 nm; and 1373, 631, 744, and 438 nm were found most sensitive for the estimation of Cu, Zn, Pb, Cr, and Cd contents in the grapevine leaves, respectively. Therefore, visible and red-edge regions were found most suitable for estimating heavy metal contents in the present study. Heavy metals played a significant role in reforming the spectral pattern of stressed grapevine compared to healthy samples, meaning that in the best structures of the SVM regression models, the concentrations of Cu, Zn, Pb, Cr, and Cd were estimated with R2 rates of 0.56, 0.85, 0.71, 0.80, and 0.86 in the testing set, respectively. The results confirm the efficiency of in-field spectroscopy in estimating heavy metals content in grapevine foliage.

Quantitative and qualitative VIS-NIR models for early determination of internal browning in ‘Cripps Pink’ apples during cold storage

Abstract ‘Cripps Pink’ apples are prone to develop internal browning disorder during cold storage rendering their commercialization particularly difficult after long-term storage. The purpose of this research was to predict internal browning defect quantitatively and qualitatively in apple, by a non-destructive equipment from spectra collected before the disorder develops. In order to obtain a broad expression of the disorder in severity and incidence, fruit treated and non-treated with 1-methylcyclopropene (1-MCP) were studied under three temperature regimes: T1) pre-cooled with forced air at −1 °C for 24 h and subsequently stored for 149 d at 0 °C; T2) placed directly at 0 °C and stored for 150 d; and T3) stored for 90 d at 5 °C then for 60 d at 0 °C. Every fruit was subjected to semi-transmittance spectral analysis between 100–1,100 nm at 0, 60, 90, 120 and 150 d of storage and matched with the presence and severity of internal browning. The disorder was quantified using image analysis in one half of cut fruit at the end of the 150 d plus 7 d at 20 °C (157 d) after verification that the damage was not evident in fruit stored before 120 d. Quantitative Support Vector Machine Regression (SVMR) model satisfactorily predicted the percentage of internal browning area per fruit shown after 157 d, as early as 90 d of storage with R2 ∼0.70, and a root mean square error for calibration (RMSEC) and prediction (RMSEP) datasets of ∼18% and ∼15% respectively. On the other hand, qualitative Partial Least Squares Discriminant Analysis (PLS-DA) model was able to predict the damaged fruit at the onset of storage (0 d) and to reach an accuracy values ∼87% in calibration and test datasets, and 12% of misclassified fruit at 90 d. Quantitative and qualitative neural network models were also evaluated, reaching similar correlation coefficients, in cross validation or accuracy rate as multivariable models did, offering a novel way for modelling biological processes. This study shows different models for predicting internal browning before the disorder appears in stored apple using semi-transmittance spectra.

Online Remaining Useful Life Prediction for Lithium-Ion Batteries Using Partial Discharge Data Features

Online accurate estimation of remaining useful life (RUL) of lithium-ion batteries is a necessary feature of any smart battery management system (BMS). In this paper, a novel partial discharge data (PDD)-based support vector machine (SVM) model is proposed for RUL prediction. The proposed algorithm extracts the critical features from the voltage and temperature of PDD to train the SVM models. The classification and regression attributes of SVM are utilized to classify and predict accurate RUL. The different ranges of PDD were analyzed to find the optimal range for training the SVM model. The SVM model trained with optimal PDD features classifies the RUL into six different classes for gross estimation, and the support vector regression is used to estimate the accurate value of the last class. The classification and predictive performance of SVM model trained using the full discharge data and PDD are compared for publicly available data. Results show that the SVM classification and regression model trained with PDD features can accurately predict the RUL with low storage pressure on BMS. The PDD-based SVM model can be utilized for online RUL estimation in electric vehicles.

Machine learning application to the reduction of ambulatory care sensitive admissions (ACSA)

Abstract Background Reducing Ambulatory Care Sensitive Admissions (ACSA) not only enhances patients’ quality of life but could also save substantial costs. ACSA are avoidable admissions for chronic conditions that are associated with socio-economic status, health status, utilization and readiness of primary care service as well as environmental factors. Undoubtedly, health authorities are highly interested in enhancing the health care services in order to reduce the number of ACSA. The objective is to identify the geographic areas where the primary care workforce should be increased in order to maximize the decrease in ACSA. Methods Using ambulatory care and inpatient claims data as well as contextual variables, we apply support vector machine regression (SVR) to select the geographic areas (fr. Bassins de vie - BVs) and the number of to-be-added primary care nurses that maximize the ACSA reduction. We also take into account the constraints related to budget and the equality of health care access. Particularly, there are three possible constraints: (1) the total number of nurses can be added in the whole region; (2) the maximum number of the nurses can be added at each area; (3) the maximum density of nurses (numbers of the nurses per 10,000 habitants) can be reached at each area. The results are visualized using spatial maps. Preliminary results In 2014, 27,000 ACSA occurred in the Occitanie, France region. For a specific set of constraints values, the model identified 16 BVs (out of 201) where the addition of 30 nurses could lead to the maximum ACSA reduction in number which is 17. Conclusions In the French Occitanie region, our SVR model was able to target a small number of geographic areas to maximize the impact of increased primary care workforce on ACSA. Our approach is applied to a single region, and it can be applied to other regions or extended at the national level as well as to other countries. Key messages A decision support tool to help health authorities in locating primary health care resources for the maximum reduction of ambulatory care sensitive admissions. An application of machine learning in primary care services.

An IoT based Intelligent Smart Energy Management System with accurate forecasting and load strategy for renewable generation

The challenge in demand side energy management lays focus on the efficient utilization of renewable sources without limiting the power consumption. To deal with the above issue, it seeks for design and development of an intelligent system with day-ahead planning and accurate forecasting of energy availability. In this work, an Intelligent Smart Energy Management Systems (ISEMS) is proposed to handle energy demand in a smart grid environment with deep penetration of renewables. The proposed scheme compares several prediction models for accurate forecasting of energy with hourly and day ahead planning. PSO based SVM regression model outperforms over several other prediction models in terms of performance accuracy. Finally, based on the predicted information, the demonstration of ISEMS experimental set-up is carried out and evaluated with different configurations considering user comfort and priority features. Also, integration of the IoT environment is developed for monitoring at the user end.

Comparison of Machine Learning Algorithms in Screening Potential Additives to Ni/Al2O3 Methanation Catalysts for Improving the Anti‐Coking Performance

AbstractIn this paper, the 16 physicochemical properties of 56 elements were processed through principal component analysis (PCA) transformation and Gaussian mixture model clustering. And then, a pool of eleven representative elements was chosen for subsequent experiments on resistance to carbon deposition. Based on the experimental results and the principal components of the selected elements, radial basis function network (RBFN), support vector machine (SVM) and Gaussian process regression (GPR) models were constructed, respectively. Compared with other models, the prediction results of GPR model are more accurate. It predicted that W element is the most effective additive, which was confirmed by further experiments. To our knowledge, this finding has not yet been formally reported in the literature. The application and implementation of this strategy provides new ideas for high dimensional, small sample and nonlinear catalyst data modeling.

Evaluating the HER-2 status of breast cancer using mammography radiomics features

PurposeThe aim of our study was to evaluate the HER-2 status in breast cancer patients using mammography (MG) radiomics features. MethodsA total of 306 Chinese female patients with invasive ductal carcinoma of no special type (IDC-NST) enrolled from January 2013 to July 2018 were divided into a training set (n = 244) and a testing set (n = 62). One hundred and eighty-six radiomics features were extracted from digital MG images based on the training set. The least absolute shrinkage and selection operator (LASSO) method was used to select the optimal predictive features for HER-2 status from the training set. Both support vector machine (SVM) and logistic regression models were employed based on the selected features. The area under the receiver operating characteristic (ROC) curves (AUCs) of the training set and testing set were used to evaluate the predictive performance of the models. ResultsCompared with the SVM model, the performance of the logistic regression model using a combination of cranial caudal (CC) and mediolateral oblique (MLO) MG views was optimal. In the training set, the sensitivity, specificity, accuracy and area under the curve (AUC) values of the logistic regression model for evaluating HER-2 status based on quantitative radiomics features were 87.29%, 58.73%, 80.00% and 0.846 (95% confidence interval (CI), 0.800-0.887), respectively, and in the testing set, the values were 73.91%, 68.75%, 77.00% and 0.787 (95% CI, 0.673-0.885), respectively. ConclusionsRadiomics features could be an efficient tool for the preoperative evaluation of HER-2 status in patients with breast cancer.

Detection of moisture content in peanut kernels using hyperspectral imaging technology coupled with chemometrics

AbstractHyperspectral imaging technology at 416–1000 nm was investigated to detect moisture content in peanut kernels. Four varieties of peanuts were scanned using a “push‐broom” system to acquire hyperspectral images. In this study, three models including partial least squares regression (PLSR), principal component regression (PCR), and support vector machine regression (SVR) were established to detect moisture content in peanut kernels based on full wavelengths. The performance of SVR was the best with determination coefficient (R2) of .9432, root mean square errors (RMSE) of 0.7054%, and residual prediction deviation (RPD) of 3.9694 for prediction set. In order to simplify modeling process and improve calculation speed of the models, successive projections algorithm (SPA) and regression coefficient were applied for optimal wavelengths selection. Then, PCR, PLSR, and SVR models were established based on these selected wavelengths, respectively. As a result, SPA–SVR generated a satisfied effect with R2 of .9363, RMSE of 0.7021%, and RPD of 3.988 for prediction set. All results in this study indicated that the combination of chemometrics and hyperspectral imaging technology could achieve rapid and nondestructive detection of moisture content in peanut kernels.Practical applicationsThe quality of peanut has a direct relationship with the moisture content in peanut. If moisture content in peanut is over standard, the peanut storage time will be shorter and also easy to be bad. It is harmful to eat this kind of peanut. Traditional methods for detection of peanut moisture content are tedious, time‐consuming, and even greatly influenced by subjective factors. As an emerging technique, hyperspectral imaging technology can detect moisture content in peanut kernels rapidly, accurately, and nondestructive. In this study, we generated a best prediction model, successive projections algorithm–support vector machine regression, which was established based on several characteristic wavelengths. All these results provided a theoretical basis for the detection of portable moisture content in peanut kernels.

Optimized prediction model for concrete dam displacement based on signal residual amendment

The traditional statistical model of concrete dam's displacement monitoring is used widely in hydraulic engineering. However, the forecasting precision of the conventional calculation model is poor due to the antiquated method of information mining and weak generalization capacity. Furthermore, the uncertain chaos effect implied in residual sequence is also intractable for modeling. In consideration of the nonlinearity, time variation, and unsteadiness of the chaotic characteristics of a dam time series, multiscale wavelet technology is used to decompose and reconstruct the residuals of multiple regression models. The fitting prediction of the low-frequency autocorrelation part is completed through the linear training ability of the autoregressive integrated moving average (ARIMA) model, and the support vector machine (SVM) regression model is constructed to optimize and process the nonlinear high-frequency signal. Then, a combined forecasting model for concrete dam's displacement based on signal residual amendment is established. The analysis of an engineering example indicates that the combined model built in this study can identify the time–frequency nonlinear characteristics of the prototype monitoring signal well, thus improving its fitting precision, antinoise ability, and robustness. In addition, the combined mathematical model established in this study is improved and developed for application to the prediction analysis of the effect quantities of other hydraulic structures.

A Container-Attachable Inertial Sensor for Real-Time Hydration Tracking.

Various sensors have been proposed to address the negative health ramifications of inadequate fluid consumption. Amongst these solutions, motion-based sensors estimate fluid intake using the characteristics of drinking kinematics. This sensing approach is complicated due to the mutual influence of both the drink volume and the current fill level on the resulting motion pattern, along with differences in biomechanics across individuals. While motion-based strategies are a promising approach due to the proliferation of inertial sensors, previous studies have been characterized by limited accuracy and substantial variability in performance across subjects. This research seeks to address these limitations for a container-attachable triaxial accelerometer sensor. Drink volume is computed using support vector machine regression models with hand-engineered features describing the container’s estimated inclination. Results are presented for a large-scale data collection consisting of 1908 drinks consumed from a refillable bottle by 84 individuals. Per-drink mean absolute percentage error is reduced by 11.05% versus previous state-of-the-art results for a single wrist-wearable inertial measurement unit (IMU) sensor assessed using a similar experimental protocol. Estimates of aggregate consumption are also improved versus previously reported results for an attachable sensor architecture. An alternative tracking approach using the fill level from which a drink is consumed is also explored herein. Fill level regression models are shown to exhibit improved accuracy and reduced inter-subject variability versus volume estimators. A technique for segmenting the entire drink motion sequence into transport and sip phases is also assessed, along with a multi-target framework for addressing the known interdependence of volume and fill level on the resulting drink motion signature.

The influence of optimized allocation of agricultural water and soil resources on irrigation and drainage in the Jingdian Irrigation District, China

Reasonable regulation of irrigation and drainage is an important way to decrease water resource waste and water pollution, and to ensure the sustainable utilization of water resources. In this study, appropriate irrigation and drainage methods are proposed by optimizing agricultural water and soil resources based on the current status of water resource utilization in irrigation district in Northwest China, and as much as possible to reduce the amount of irrigation and drainage while ensuring the necessary for salt leaching. A two-layer model was considered to maximize economic benefits and relative production through a nonlinear algorithm to optimize the crop acreage, irrigation quota and the amount of irrigation water at each crop growth stage. A support vector machine regression model for predicting drainage was constructed, including drainage and irrigation, precipitation, evaporation and groundwater depth. Additionally, the amount of drainage water was compared before and after the optimization of irrigation. The amount of irrigation water demand in a wet year (2014), normal year (2008) and dry year (2013) decreased by 23.85 million m3, 12.85 million m3 and 17.50 million m3, respectively, after optimization of the crop planting structure and irrigation scheduling. Furthermore, the net economic benefit increased by 3.76 billion yuan, 1.14 billion yuan and 2.34 billion yuan, as compared with the actual output value. The amount of drainage water decreased by 8.60%, 8.93% and 5.21% compared to that with no optimization in a wet year (2014), normal year (2008) and dry year (2013), respectively. This method can scientifically allocate the amount of irrigation and drainage, feasibly improve economic benefits and prevent soil salinization, thus providing guidance for efficient water resource utilization and ecological protection in arid areas.

Combination of Multi-Temporal Sentinel 2 Images and Aerial Image Based Canopy Height Models for Timber Volume Modelling

Multi-temporal Sentinel 2 optical images and 3D photogrammetric point clouds can be combined to enhance the accuracy of timber volume models on large spatial scale. Information on the proportion of broadleaf and conifer trees improves timber volume models obtained from 3D photogrammetric point clouds. However, the broadleaf-conifer information cannot be obtained from photogrammetric point clouds alone. Furthermore, spectral information of aerial images is too inconsistent to be used for automatic broadleaf-conifer classification over larger areas. In this study we combined multi-temporal Sentinel 2 optical satellite images, 3D photogrammetric point clouds from digital aerial stereo photographs, and forest inventory plots representing an area of 35,751 km2 in south-west Germany for (1) modelling the percentage of broadleaf tree volume (BL%) using Sentinel 2 time series and (2) modelling timber volume per hectare using 3D photogrammetric point clouds. Forest inventory plots were surveyed in the same years and regions as stereo photographs were acquired (2013–2017), resulting in 11,554 plots. Sentinel 2 images from 2016 and 2017 were corrected for topographic and atmospheric influences and combined with the same forest inventory plots. Spectral variables from corrected multi-temporal Sentinel 2 images were calculated, and Support Vector Machine (SVM) regressions were fitted for each Sentinel 2 scene estimating the BL% for corresponding inventory plots. Variables from the photogrammetric point clouds were calculated for each inventory plot and a non-linear regression model predicting timber volume per hectare was fitted. Each SVM regression and the timber volume model were evaluated using ten-fold cross-validation (CV). The SVM regression models estimating the BL% per Sentinel 2 scene achieved overall accuracies of 68%–75% and a Root Mean Squared Error (RMSE) of 21.5–26.1. The timber volume model showed a RMSE% of 31.7%, a mean bias of 0.2%, and a pseudo-R2 of 0.64. Application of the SVM regressions on Sentinel 2 scenes covering the state of Baden-Württemberg resulted in predictions of broadleaf tree percentages for the entire state. These predicted values were used as additional predictor in the timber volume model, allowing for predictions of timber volume for the same area. Spatially high-resolution information about growing stock is of great practical relevance for forest management planning, especially when the timber volume of a smaller unit is of interest, for example of a forest stand or a forest district where not enough terrestrial inventory plots are available to make reliable estimations. Here, predictions from remote-sensing based models can be used. Furthermore, information about broadleaf and conifer trees improves timber volume models and reduces model errors and, thereby, prediction uncertainties.

Evaluation of data-driven models (SVR and ANN) for groundwater-level prediction in confined and unconfined systems

Modeling the behavior of groundwater levels is necessary to implement sustainable groundwater resource management. Groundwater is a non-linear and complex system, which can be modeled by data-driven models. This study evaluates the performances of data-driven models, support vector machine regression (SVR) and artificial neural network (ANN), for forecasting groundwater levels of confined and unconfined systems at 1-, 2-, and 3-month ahead. This is the first time that confined and unconfined aquifers have been compared using data-driven models. In addition, to identify the optimal input combination, a hybrid gamma test (GT) and genetic algorithm (GA) was used. The coefficient of correlation (R), Mean Absolute Error (MAE), root mean squared error (RMSE), Nash–Sutcliffe efficiency (NSE), and developed discrepancy ratio (DDR) were applied to evaluate the SVR and ANN models. Results showed that the SVR and ANN models were more accurate for the unconfined system than the confined system for forecasts up to 3-month ahead. In both hydrogeological systems for 1-month ahead, the models performed better than for 2- and 3-month ahead forecasts, and the accuracy of the models decreased as the months ahead increased. The SVR model performed better than the ANN model for 1-, 2-, and 3-month ahead groundwater-level forecasting. The SVR model could be successfully used in predicting monthly groundwater in confined and unconfined systems.