Youngmin Seo

Modeling Nonlinear Monthly Evapotranspiration Using Soft Computing and Data Reconstruction Techniques

The objective of this study is to develop soft computing and data reconstruction techniques for modeling monthly California Irrigation Management Information System (CIMIS) evapotranspiration (ETo) at two stations, U.C. Riverside and Durham, in California. The nonlinear dynamics of monthly CIMIS ETo is examined using autocorrelation function, phase space reconstruction, and close returns plot. The generalized regression neural networks and genetic algorithm (GRNN-GA) conjunction model is developed for modeling monthly CIMIS ETo. Among different input variables considered, solar radiation (RAD) is found to be the most effective variable for modeling monthly CIMIS ETo using GRNN-GA for both stations. Adding other input variables to the best 1-input combination improves the model performance. The generalized regression neural networks and backpropagation algorithm (GRNN-BP) conjunction model is compared with the results of GRNN-GA for modeling monthly CIMIS ETo. Two bootstrap resampling methods are implemented to reconstruct the training data. Method 1 (1-BGRNN-GA) employs simple extensions of training data using the bootstrap resampling method. For each training data, method 2 (2-BGRNN-GA) uses individual bootstrap resampling of original training data. Results indicate that Method 2 (2-BGRNN-GA) improves modeling of monthly CIMIS ETo and is more stable and reliable than are GRNN-GA, GRNN-BP, and Method 1 (1-BGRNN-GA).

Assessment of Pan Evaporation Modeling Using Bootstrap Resampling and Soft Computing Methods

AbstractThis study evaluated combined bootstrap resampling and neural network models for estimating daily pan evaporation (PE) in the Republic of Korea. Two different support vector machines (SVMs), e-support vector regression (e-SVR) and ν-support vector regression (ν-SVR), were developed for the local implementation of SVMs. Five-input combination models (e-SVR5 and ν-SVR5) were found to be generally the best for the local implementation of SVMs. Optimal SVMs, including e-SVR5 and ν-SVR5, were employed to develop bootstrap-based support vector machines (BSVMs) for two weather stations. The ensemble PE was estimated by averaging the output of 50 individual BSVMs. A Mann-Whitney U test was performed to compare the observed and ensemble of bootstrap resamplings for the training data of the PE. The uncertainty associated with PE estimation using BSVMs was evaluated. Results indicated that BSVMs could improve confidence in PE estimation and that ensemble PE using BSVMs is more stable and reliable than that u...

Multi-spatial-frequency and phase-shifting profilometry using a liquid crystal phase modulator

Optical profilometry is widely applied for measuring the morphology of objects by projecting predetermined patterns on them. In this technique, the compact size is one of the interesting issues for practical applications. The generation of pattern by the interference of coherent light sources has a potential to reduce the dimension of the illumination part. Moreover, this method can make fine patterns without projection optics, and the illumination part is free of restriction from the numerical aperture of the projection optics. In this paper, a phase-shifting profilometry is implemented by using a single liquid crystal (LC) cell. The LC phase modulator is designed to generate the interference patterns with several different spatial frequencies by changing selection of the spacing between the micro-pinholes. We manufactured the LC phase modulator and calibrated it by measuring the phase modulation amount depending on an applied voltage. Our optical profilometry using the single LC cell can generate multi-spatial frequency patterns as well as four steps of the phase-shifted patterns. This method can be implemented compactly, and the reconstructed depth profile is obtained without a phase-unwrapping algorithm.

Evaluation of pan evaporation modeling with two different neural networks and weather station data

This study evaluates neural networks models for estimating daily pan evaporation for inland and coastal stations in Republic of Korea. A multilayer perceptron neural networks model (MLP-NNM) and a cascade correlation neural networks model (CCNNM) are developed for local implementation. Five-input models (MLP 5 and CCNNM 5) are generally found to be the best for local implementation. The optimal neural networks models, including MLP 4, MLP 5, CCNNM 4, and CCNNM 5, perform well for homogeneous (cross-stations 1 and 2) and nonhomogeneous (cross-stations 3 and 4) weather stations. Statistical results of CCNNM are better than those of MLP-NNM during the test period for homogeneous and nonhomogeneous weather stations except for MLP 4 being better in BUS-DAE and POH-DAE, and MLP 5 being better in POH-DAE. Applying the conventional models for the test period, it is found that neural networks models perform better than the conventional models for local, homogeneous, and nonhomogeneous weather stations.

Estimating Spatial Precipitation Using Regression Kriging and Artificial Neural Network Residual Kriging (RKNNRK) Hybrid Approach

A hybrid model, combining regression kriging and neural network residual kriging (RKNNRK), is developed for determining spatial precipitation distribution. The RKNNRK model is compared with current spatial interpolation models, including simple kriging (SK), ordinary kriging (OK), universal kriging (UK), regression kriging (RK) and neural network residual kriging (NNRK). Results show that hybrid models, including RK, NNRK and RKNNRK, performed better than SK, OK and UK, based on the coefficient of efficiency (CE), coefficient of determination (r2), index of agreement (d), mean squared relative error (MSRE), mean absolute error (MAE), root-mean-square error (RMSE), and mean squared error (MSE). Of the six spatial interpolation models, the RKNNRK model was the most accurate, and the NNRK model was the second best.

Comparison of different heuristic and decomposition techniques for river stage modeling

This paper proposes hybrid soft computing models for daily river stage modeling. The models combine variational mode decomposition (VMD) with different soft computing models, including artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), and random forest (RF). The performances of VMD-based models (VMD-ANN, VMD-ANFIS, and VMD-RF) are assessed by model efficiency indices and graphical comparison, and compared with those of single models (ANN, ANFIS, and RF) and ensemble empirical mode decomposition (EEMD)-based models (EEMD-ANN, EEMD-ANFIS, and EEMD-RF). Results show that VMD-ANN, VMD-ANFIS, and VMD-RF models are more efficient and accurate than ANN, ANFIS, and RF models, respectively, and slightly better than EEMD-ANN, EEMD-ANFIS, and EEMD-RF models, respectively. In terms of model efficiency and accuracy, the top five models are VMD-ANFIS, EEMD-ANFIS, VMD-ANN, VMD-RF, and ANFIS and the VMD-ANFIS model is the best. It is found that VMD can enhance the performance of conventional single soft computing models; VMD is more effective than EEMD for hybrid model development; and the ANFIS model combined with VMD and EEMD can yield better efficiency and accuracy than other models. Therefore, VMD-based hybrid modeling is a more effective method for reliable daily river stage modeling.

Computation of Daily Solar Radiation Using Wavelet and Support Vector Machines: A Case Study

The objective of this study is to apply a hybrid model for estimating solar radiation and investigate its accuracy. A hybrid model is wavelet-based support vector machines (WSVMs). Wavelet decomposition is employed to decompose the solar radiation time series components into approximation and detail components. These decomposed time series are then used as input of support vector machines (SVMs) modules in the WSVMs model. Based on statistical indexes, results indicate that WSVMs can successfully be used for the estimation of daily global solar radiation at Champaign and Springfield stations in Illinois.

River Stage Forecasting Model Combining Wavelet Packet Transform and Artificial Neural Network

A reliable streamflow forecasting is essential for flood disaster prevention, reservoir operation, water supply and water resources management. This study proposes a hybrid model for river stage forecasting and investigates its accuracy. The proposed model is the wavelet packet-based artificial neural network(WPANN). Wavelet packet transform(WPT) module in WPANN model is employed to decompose an input time series into approximation and detail components. The decomposed time series are then used as inputs of artificial neural network(ANN) module in WPANN model. Based on model performance indexes, WPANN models are found to produce better efficiency than ANN model. WPANN-sym10 model yields the best performance among all other models. It is found that WPT improves the accuracy of ANN model. The results obtained from this study indicate that the conjunction of WPT and ANN can improve the efficiency of ANN model and can be a potential tool for forecasting river stage more accurately.

Uncertainty Analysis of Parameters of Spatial Statistical Model Using Bayesian Method for Estimating Spatial Distribution of Probability Rainfall

This study applied the Bayesian method for the quantification of the parameter uncertainty of spatial linear mixed model in the estimation of the spatial distribution of probability rainfall. In the application of Bayesian method, the prior sensitivity analysis was implemented by using the priors normally selected in the existing studies which applied the Bayesian method for the puppose of assessing the influence which the selection of the priors of model parameters had on posteriors. As a result, the posteriors of parameters were differently estimated which priors were selected, and then in the case of the prior combination, F-S-E, the sizes of uncertainty intervals were minimum and the modes, means and medians of the posteriors were similar to the estimates using the existing classical methods. From the comparitive analysis between Bayesian and plug-in spatial predictions, we could find that the uncertainty of plug-in prediction could be slightly underestimated than that of Bayesian prediction.

Evaluation of daily solar radiation flux using soft computing approaches based on different meteorological information: peninsula vs continent

This study compares single and hybrid soft computing models for estimating daily solar radiation flux for two scenarios. Scenario I developed single soft computing models, including multilayer perceptron (MLP), support vector machines (SVM), adaptive neuro-fuzzy inference system (ANFIS), and multivariate adaptive regression splines (MARS), for estimating daily solar radiation flux at two stations from the USA and South Korea. The MLP model was used to evaluate the effect of factors controlling daily solar radiation flux. Using different combinations of controlling factors as input, the MLP and SVM models, based on evaluation measures, were found to be superior to the ANFIS and MARS models at Big Bend station, USA. In addition, the MLP, SVM, and MARS models performed better than did the ANFIS model at Incheon station, South Korea. Scenario II combined the discrete wavelet transform (DWT) and single soft computing models (e.g., MLP and SVM) for improved performance using 4-input combination. The wavelet-based MLP (WMLP) and SVM (WSVM) models were superior to other single soft computing models (MLP, SVM, ANFIS, and MARS) at two stations. Taylor diagrams, violin plots and point density plots were also utilized to examine the similarity between the observed and estimated solar radiation flux values. Results showed that scenarios I and II can be alternatives for estimating daily solar radiation flux based on different meteorological information, such as peninsular and continental conditions.