Search

This paper is aimed to develop a hybrid model (PCA-ANN) to predict the extractive content in eucalyptus wood clones. The input variables to the virtual analyzer are planting parameters available from the forest inventory, therefore the prediction does not rely on laboratory analysis of the wood samples, affording a quick estimate of the extractive contents. This study further bridges the literature gap on the investigation of the cause of variability of extractive content in eucalyptus wood. The PCA-ANN was identified from experimental data to predict and monitor the extractive content, since laboratory measurements can take several days and become available only after wood processing. The experimental data contained information on ten species of eucalyptus clones from five regions in the extreme south of Bahia, Brazil. Principal Component Analysis (PCA) firstly assessed the impact of planting variables on the extractives content. The variability of the data was represented by eight principal components and the variables that mostly contribute to the extractive content are: potential acidity, iron, saturation of aluminum, magnesium, pH, base saturation, remaining phosphorus, zinc, manganese and copper. The artificial neural network (ANN) with the 8 principal components in the input layer showed that the PCA could effectively reduce the dimensionality of the data. For practical purposes, though, the ANN with 10 input variables and 16 neurons in the hidden layer, presenting an average relative deviation of 1.5%, is recommended. The prediction of the extractive content is essential to allow preventive management practices toward the improvement of yield and quality of the cellulosic pulp. [ABSTRACT FROM AUTHOR]

This paper has as a main goal to evaluate how models of the forecast will work with a group of variables that were selected based only on their correlation with the average tariff variation. Two forecast models are used, the first based on multiple linear regression techniques and the second based on the application of artificial neural networks (perceptron). We intend to use those models to reach the current water tariff based on the historic variation of the charge and the selected variables applied to municipal and private companies that operate water supply and wastewater systems in the South and Southeast regions of Brazil. The subsidiary data for the elaboration of the models were obtained through the National Sanitation Information System (SNIS). The obtained results indicated that the forecasting processes, in both models used, were able to forecast with high accuracy the fees, and guaranteed the maintenance of the surplus for the analyzed systems. [ABSTRACT FROM AUTHOR]

Empirical models based on machine learning methods have been used for landslide susceptibility mapping. The most accurate model is usually chosen to generate the final map. This paper demonstrates the importance of analyzing the spatial pattern of susceptibility maps, since models with similar performance can produce different output values. The relevance of terrain attributes and the sensitivity of models to input variables are also discussed. The applications of random forest (RF) and artificial neural network (ANN) models to the identification of landslide susceptible areas in the Fão River Basin, Southern Brazil, were evaluated and compared. The following have been included in the methodology: (1) the extraction of predictive attributes (e.g., slope, aspect, curvatures, valley depth) from a digital elevation model; (2) the organization of a landslide scar inventory; (3) the calibration and validation procedures of the models; (4) the analysis of model performance according to accuracy (area under the receiver operating characteristic curve) and parsimony (Akaike Information Criterion); (5) the reclassification of maps into susceptibility categories. All model configurations resulted in an accuracy above 0.9, demonstrating the ability of both techniques in landslide susceptibility mapping. The RF model stood out in this respect, recording the highest accuracy index among all tested configurations (0.949). The ANN model was more parsimonious, obtaining an accuracy of 0.925 with a much smaller number of internal connections. Thus, even with both having high and equivalent accuracy indexes, the models can establish different relationships between the input and the output susceptibility indexes, resulting in various possible landslide occurrence scenarios. These differences, together with the difficulty in defining which model presents more coherent results, reinforce the possibility of extracting spatial statistics, considering multiple configurations of models that combine accuracy and parsimony, in landslide susceptibility mapping. [ABSTRACT FROM AUTHOR]

Precipitation, in short periods of time, is a phenomenon associated with high levels of uncertainty and variability. Given its nature, traditional forecasting techniques are expensive and computationally demanding. This paper presents a soft computing technique to forecast the occurrence of rainfall in short ranges of time by artificial neural networks (ANNs) in accumulated periods from 3 to 7 days for each climatic season, mitigating the necessity of predicting its amount. With this premise it is intended to reduce the variance, rise the bias of data and lower the responsibility of the model acting as a filter for quantitative models by removing subsequent occurrences of zeros values of rainfall which leads to bias the and reduces its performance. The model were developed with time series from ten agriculturally relevant regions in Brazil, these places are the ones with the longest available weather time series and and more deficient in accurate climate predictions, it was available 60 years of daily mean air temperature and accumulated precipitation which were used to estimate the potential evapotranspiration and water balance; these were the variables used as inputs for the ANNs models. The mean accuracy of the model for all the accumulated periods were 78% on summer, 71% on winter 62% on spring and 56% on autumn, it was identified that the effect of continentality, the effect of altitude and the volume of normal precipitation, have an direct impact on the accuracy of the ANNs. The models have peak performance in well defined seasons, but looses its accuracy in transitional seasons and places under influence of macro-climatic and mesoclimatic effects, which indicates that this technique can be used to indicate the eminence of rainfall with some limitations. [ABSTRACT FROM AUTHOR]

This study assesses the deterministic and probabilistic forecasting skill of a 1-month-lead ensemble of Artificial Neural Networks (EANN) based on low-frequency climate oscillation indices. The predictand is the February-April (FMA) rainfall in the Brazilian state of Ceará, which is a prominent subject in climate forecasting studies due to its high seasonal predictability. Additionally, the study proposes combining the EANN with dynamical models into a hybrid multi-model ensemble (MME). The forecast verification is carried out through a leave-one-out cross-validation based on 40 years of data. The EANN forecasting skill is compared with traditional statistical models and the dynamical models that compose Ceará's operational seasonal forecasting system. A spatial comparison showed that the EANN was among the models with the smallest Root Mean Squared Error (RMSE) and Ranked Probability Score (RPS) in most regions. Moreover, the analysis of the area-aggregated reliability showed that the EANN is better calibrated than the individual dynamical models and has better resolution than Multinomial Logistic Regression for above-normal (AN) and below-normal (BN) categories. It is also shown that combining the EANN and dynamical models into a hybrid MME reduces the overconfidence of the extreme categories observed in a dynamically-based MME, improving the reliability of the forecasting system. [ABSTRACT FROM AUTHOR]

The interpretation of lung auscultation is highly subjective and relies on non-specific nomenclature. Computer-aided analysis has the potential to better standardize and automate evaluation. We used 35.9 hours of auscultation audio from 572 pediatric outpatients to develop DeepBreath : a deep learning model identifying the audible signatures of acute respiratory illness in children. It comprises a convolutional neural network followed by a logistic regression classifier, aggregating estimates on recordings from eight thoracic sites into a single prediction at the patient-level. Patients were either healthy controls (29%) or had one of three acute respiratory illnesses (71%) including pneumonia, wheezing disorders (bronchitis/asthma), and bronchiolitis). To ensure objective estimates on model generalisability, DeepBreath is trained on patients from two countries (Switzerland, Brazil), and results are reported on an internal 5-fold cross-validation as well as externally validated (extval) on three other countries (Senegal, Cameroon, Morocco). DeepBreath differentiated healthy and pathological breathing with an Area Under the Receiver-Operator Characteristic (AUROC) of 0.93 (standard deviation [SD] ± 0.01 on internal validation). Similarly promising results were obtained for pneumonia (AUROC 0.75 ± 0.10), wheezing disorders (AUROC 0.91 ± 0.03), and bronchiolitis (AUROC 0.94 ± 0.02). Extval AUROCs were 0.89, 0.74, 0.74 and 0.87 respectively. All either matched or were significant improvements on a clinical baseline model using age and respiratory rate. Temporal attention showed clear alignment between model prediction and independently annotated respiratory cycles, providing evidence that DeepBreath extracts physiologically meaningful representations. DeepBreath provides a framework for interpretable deep learning to identify the objective audio signatures of respiratory pathology. [ABSTRACT FROM AUTHOR]

Accurate simulations of significant wave height (Hs) are extremely important for the safety of navigation, port operations, and oil and gas exploration. Thus, accurate forecasts of Hs are essential for accident prevention and maintenance of services vital to the economy. Considering the limitations of traditional numerical modeling, such as the typical model underestimation of Hs under severe conditions, forecasting Hs using artificial neural networks is a promising method and a complementary approach. In this study we develop a post-processing model using Long Short-Term Memory (LSTM) algorithm to improve outputs from the numerical model WAVEWATCH III (WW3) at Santos Basin, Brazil. The hybrid scheme is focused on the simulations of 1-, 2-, 3- and 4-day residues (difference between observations and WW3) using measurements from a local wave buoy moored in deep water. The results of the hybrid model (WW3+LSTM) show a better performance compared with WW3, being capable of better representing the peak of the events and storms. On average, the gains from using WW3+LSTM reach 3.8% in Correlation Coefficient (CORR), 14.2% in Bias (BIAS), 10.2% in Root Mean Squared Error (RMSE), and 10.7% in Scatter Index (SI). The hybrid model developed allows high-skill forecasts to be carried out on large domains and through longer horizons. [ABSTRACT FROM AUTHOR]

This study aimed to assess monthly average daily global solar irradiation ( H g m) estimates in Southeastern Brazil from empirical models and artificial intelligence (AI) techniques using extreme air temperatures (maximum—Tx and minimum—Tn) and extraterrestrial solar irradiation ( H 0 m ) and evaluate the geographic and climate factors in the performance of the models or the AI techniques. The H g m , Tx and Tn data series from 95 automatic weather stations of the Brazilian National Institute of Meteorology (INMET) were used. The Hargreaves-Samani (HS) and Bristow-Campbell (BC) models were fitted, and AI techniques (artificial neural network—ANN and support vector machines—SVM) were trained, and assessed according to local climate conditions. Three input schemes in the AI techniques were assessed: (i) H 0 m , Tx, and Tn; (ii) H 0 m and air temperature amplitude (ΔT); and (iii) H 0 m , Tx, Tn, and ΔT. The coefficients of the HS (0.134 ≤ kr ≤ 0.262) and BC (0.539 ≤ β0 ≤ 0.796; 0.004 ≤ β1 ≤ 0.843; 0.35 ≤ β2 ≤ 2.60) models and their performance were influenced by climate, continentality/maritime effects, altitude and weather systems. The performance of the HS model was inferior to that of BC model and the SVM and ANN techniques. Overall, the SVN and ANN techniques performed better than the BC method. However, BC had similar or superior precision and accuracy to SVM and ANN for some climatic conditions and combinations of input variables. The quality of the H g m series, the combination of input variables, and climate influenced the estimates of the SVM and ANN techniques. [ABSTRACT FROM AUTHOR]

In this paper, laser-induced breakdown spectroscopy (LIBS) combined with artificial neural network (ANN) was investigated to classify four species of wood samples (Africa rosewood, Brazil bubinga, Myanmar padauk, and Pterocarpus erinaceus). The wood samples were ablated by laser pulses to generate plasma emission, which was measured by a spectrometer and transmitted into a computer for further data analysis. The feature spectral data were selected out based on loadings of principal component analysis (PCA) and normalized using the sum of all feature spectra data. The ANN model was built based on the feature spectral data to classify the wood species. The relationship between correct classification rate (CCR) and settings of ANN was discussed. The CCR of ANN model for test set data achieved 100% with multilayer perceptron network and Broyden–Fletcher–Goldfarb–Shanno iterative algorithm. This result was also compared with the CCRs of PLS-DA, KNN, and SIMCA model for test set (82.5%, 95.83%, and 51.67%, respectively). Using the ratio between feature variables to recognize the species of wood was also discussed. The experimental results demonstrated that LIBS integrated with ANN could be applied for analyzing and recognizing wood species. [ABSTRACT FROM AUTHOR]

Prediction of plantation productivity from environmental data is challenged by the complex relationships defining the growth and yield processes. Artificial neural network (ANN) can be used to model complex nonlinear interrelations but are challenged by the choice of the training algorithm and the structure and size of the network. The objective of the present study is to find an efficient ANN to estimate the eucalyptus productivity from biotic, abiotic and silvicultural data. We defined the efficiency of an ANN as the processing time to supply an accurate solution. We increase the efficiency of the network in two steps: one outside the ANN, through principal component analysis (PCA), and one inside the ANN, through dedicated pruning methods. To test different network configurations, we used data from 507 Eucalyptus plantations, 3 to 7 years old, located in Minas Gerais, Brazil. An ANN configuration is a combination of the number of neurons in the hidden layers, training algorithm and pruning method. Each ANN was trained five times to assess the impact of the initial weights on the results, which led to a factorial experiment with 9000 combinations. The most accurate result was supplied in 38.81 s by an ANN using the data trimmed with PCA, trained with the Scaled Conjugate Gradient algorithm with four neurons and Magnitude-Based Pruning method. However, an accuracy loss of less than 1% (i.e., the second most accurate result) was obtained in 1.7 s using the same ANN configuration, but no pruning. Our results indicate that an efficient prediction of Eucalyptus productivity does not use all the data or the most complex training algorithms. The proposed efficient ANN supplies the most precise results compared with the existing models. Furthermore, the efficient ANN produces more accurate results from environmental variables than some models that contain dimensional variables. Our study suggests that modeling the yield of Eucalyptus plantations should be executed using a large set of environmental variables, as no clear edaphic or climatic variables supplied accurate results. [ABSTRACT FROM AUTHOR]

Two Artificial Intelligence (AI) methods, Fuzzy Inference System (FIS) and Artificial Neural Network (ANN), are applied to Landslide Susceptibility Mapping (LSM), to compare complementary aspects of the potentials of the two methods and to extract physical relationships from data. An index is proposed in order to rank and filter the FIS rules, selecting a certain number of readable rules for further interpretation of the physical relationships among variables. The area of study is Rolante river basin, southern Brazil. Eleven attributes are generated from a Digital Elevation Model (DEM), and landslide scars from an extreme rainfall event are used. Average accuracy and area under Receiver Operating Characteristic curve (AUC) resulted, respectively, in 81.27% and 0.8886 for FIS, and 89.45% and 0.9409 for ANN. ANN provides a map with more amplitude of outputs and less area classified as high susceptibility. Among the 40 (10%) best-ranked FIS rules, 13 have high susceptibility output, while 27 have low; a cause is that low susceptibility areas are larger on the map. Slope is highly connected to susceptibility. Elevation, when high (plateau) or low (floodplain), inhibits high susceptibility. Six attributes show the same fuzzy set for the 18 best-ranked rules, meaning this fuzzy set is common on the map. Overall findings point out that ANN is best suited for LSM map generation, but, based on them, using FIS is important to help researchers understand more about AI models for LSM and about landslide phenomenon. [ABSTRACT FROM AUTHOR]

Water quality monitoring is a complex issue that requires support tools in order to provide information for water resource management. Budget constraints as well as an inadequate water quality network design call for the development of evaluation tools to provide efficient water quality monitoring. For this purpose, a nonlinear principal component analysis (NLPCA) based on an autoassociative neural network was performed to assess the redundancy of the parameters and monitoring locations of the water quality network in the Piabanha River watershed. Oftentimes, a small number of variables contain the most relevant information, while the others add little or no interpretation to the variability of water quality. Principal component analysis (PCA) is widely used for this purpose. However, conventional PCA is not able to capture the nonlinearities of water quality data, while neural networks can represent those nonlinear relationships. The results presented in this work demonstrate that NLPCA performs better than PCA in the reconstruction of the water quality data of Piabanha watershed, explaining most of data variance. From the results of NLPCA, the most relevant water quality parameter is fecal coliforms (FCs) and the least relevant is chemical oxygen demand (COD). Regarding the monitoring locations, the most relevant is Poço Tarzan (PT) and the least is Parque Petrópolis (PP). [ABSTRACT FROM AUTHOR]

Purpose: We investigated the application of Kohonen Neural Networks (KNNs) in order to estimate sediment yield based on runoff and climatological data in a semiarid region of Brazil. Accurate estimations of sediment yield are essential to improve the management of soil erosion in semiarid areas, where large quantities of sediments tend to be produced only periodically. Materials and methods: The case study is an erosion plot within the São João do Cariri Experimental Basin, which is located in the semiarid portion of Paraíba State, Brazil. KNNs are unsupervised neural networks capable of reducing a multidimensional data set to a bidimensional matrix of features, which can be used for analysis and prediction purposes. A total of 60 rainfall events, which occurred between 1999 and 2002, were used to calibrate and test the model. The application of a multivariate linear regression (MLR) model was also carried out. Results and discussion: Statistical indexes were used as criteria for evaluating the performance of the KNN and MLR models for the test data set. The correlation and relative bias of the KNN model estimations with those from observed data were 0.90 and −4.39 %, respectively. A correlation of 0.70 and a relative bias of 15.63 % were found from the comparison of sediment yields obtained by the MLR model with those of the observed data. Analysis of the outcomes indicates that the KNN model, which is capable of detecting and extracting nonlinear trends, produced more reliable results than the regression model. Conclusions: The KNN model results appear to be superior to those generated by the MLR model and suggest that the developed methodology may be applied to similar case studies. [ABSTRACT FROM AUTHOR]

The water quality index (WQI) is an important tool for water resource management and planning. However, it has major disadvantages: the generation of chemical waste, is costly, and time-consuming. In order to overcome these drawbacks, we propose to simplify this index determination by replacing traditional analytical methods with ultraviolet-visible (UV-Vis) spectrophotometry associated with artificial neural network (ANN). A total of 100 water samples were collected from two rivers located in Assis, SP, Brazil and calculated the WQI by the conventional method. UV-Vis spectral analyses between 190 and 800 nm were also performed for each sample followed by principal component analysis (PCA) aiming to reduce the number of variables. The scores of the principal components were used as input to calibrate a three-layer feed-forward neural network. Output layer was defined by the WQI values. The modeling efforts showed that the optimal ANN architecture was 19-16-1, trainlm as training function, root-mean-square error (RMSE) 0.5813, determination coefficient between observed and predicted values (R2) of 0.9857 (p < 0.0001), and mean absolute percentage error (MAPE) of 0.57% ± 0.51%. The implications of this work’s results open up the possibility to use a portable UV-Vis spectrophotometer connected to a computer to predict the WQI in places where there is no required infrastructure to determine the WQI by the conventional method as well as to monitor water body’s in real time. [ABSTRACT FROM AUTHOR]