Semi-empirical prediction method for monthly precipitation prediction based on environmental factors and comparison with stochastic and machine learning models

ABSTRACT

Precipitation prediction is central in hydrology and water resources planning and management. This paper introduces a semi-empirical predictive model to predict monthly precipitation and compares its predictive skill with those of machine learning (ML) methods. The stochastic method presented herein estimates monthly precipitation with one-step-ahead prediction properties. The ML predictive skill of the algorithms is evaluated by predicting monthly precipitation relying on the statistical association between precipitation and environmental and topographic factors. The semi-empirical predictive model features non-negative matrix factorization (NMF) for investigating the influence of multiple predictor variables on precipitation. The semi-empirical predictive model’s parameters are optimized with the hybrid genetic algorithm (GA) and Levenberg-Marquardt algorithm (LM), or GALMA, yielding a validated model with high predictive skill. The methodologies are illustrated with data from Hubei Province, China, which comprise 27 meteorological station datasets from 1988–2017. The empirical results provide valuable insights for developing semi-empirical rainfall prediction models.     

基于小波神经网络耦合模型对云南星云湖富营养化气象驱动因子的分析

气象因子是影响湖泊富营养化的重要因素,而湖泊富营养化对人群健康、生态系统和社会经济等均有负面影响.本文基于统计资料及遥感数据,结合Morlet小波分析和BP多层前馈神经网络(BP神经网络)构建了不同时间尺度下的小波神经网络耦合模型,分析了19862011年云南星云湖水华强度变化与月降雨量、月平均气温、月平均风速、月日照...     

Prediction of hourly actual evapotranspiration using neural networks,genetic programming,and statistical models

The complexity of the evapotranspiration process and its variability in time and space have imposed some limitations on previously developed evapotranspiration models. In this study, two data‐driven models: genetic programming (GP) and artificial neural networks (ANNs), and statistical regression models were developed and compared for estimating the hourly eddy covariance (EC)‐measured actual evapotranspiration (AET) using meteorological variables. The utility of the investigated data‐driven models was also compared with that of HYDRUS‐1D model, which makes use of conventional Penman–Monteith (PM) model for the prediction of AET. The latent heat (LE), which is measured using the EC method, is modelled as a function of five climatic variables: net radiation, ground temperature, air temperature, relative humidity, and wind speed in a reconstructed landscape located in Northern Alberta, Canada. Several ANN models were evaluated using two training algorithms of Levenberg–Marquardt and Bayesian regularization. The GP technique was used to generate mathematical equations correlating AET to the five climatic variables. Furthermore, the climatic variables, as well as their two‐factor interactions, were statistically analysed to obtain a regression equation and to indicate the climatic factors having significant effect on the evapotranspiration process. HYDRUS‐1D model as an available physically based model was examined for estimating AET using climatic variables, leaf area index (LAI), and soil moisture information. The results indicated that all three proposed data‐driven models were able to approximate the AET reasonably well; however, GP and regression models had better generalization ability than the ANN model. The results of HYDRUS‐1D model exhibited that a physically based model, such as HYDRUS‐1D, might be comparable or even inferior to the data‐driven models in terms of the overall prediction accuracy. Based on the developed GP and regression models, net radiation and ground temperature had larger contribution to the AET process than other variables. Copyright © 2010 John Wiley & Sons, Ltd.     

Modeling and Prediction of Monthly Total Ozone Concentrations by Use of an Artificial Neural Network Based on Principal Component Analysis

In the work discussed in this paper we considered total ozone time series over Kolkata (22°34′10.92″N, 88°22′10.92″E), an urban area in eastern India. Using cloud cover, average temperature, and rainfall as the predictors, we developed an artificial neural network, in the form of a multilayer perceptron with sigmoid non-linearity, for prediction of monthly total ozone concentrations from values of the predictors in previous months. We also estimated total ozone from values of the predictors in the same month. Before development of the neural network model we removed multicollinearity by means of principal component analysis. On the basis of the variables extracted by principal component analysis, we developed three artificial neural network models. By rigorous statistical assessment it was found that cloud cover and rainfall can act as good predictors for monthly total ozone when they are considered as the set of input variables for the neural network model constructed in the form of a multilayer perceptron. In general, the artificial neural network has good potential for predicting and estimating monthly total ozone on the basis of the meteorological predictors. It was further observed that during pre-monsoon and winter seasons, the proposed models perform better than during and after the monsoon.     

Comparison of daily streamflow forecasts using extreme learning machines and the random forest method

ABSTRACT

Daily streamflow forecasting is a challenging and essential task for water resource management. The main goal of this study was to compare the accuracy of five data-driven models: extreme learning machine (basic ELM), extreme learning machine with kernels (ELM-kernel), random forest (RF), back-propagation neural network (BPNN) and support vector machine (SVR). The results show that the ELM-kernel model provided a superior alternative to the other models, and the basic ELM model had the poorest performance. To further evaluate the predictive capacities of the five models, the estimations of low flow and high flow in the testing dataset were compared. The RF model was slightly superior to the other models in predicting the peak flows, and the ELM-kernel model showed the highest prediction precision of low flows. There was no single model that showed obvious advantages over the other models in this study. Therefore, further exploration is required for the hydrological forecasting problems.     

A stepwise-cluster forecasting approach for monthly streamflows based on climate teleconnections

In this study, a stepwise cluster forecasting (SCF) framework is proposed for monthly streamflow prediction in Xiangxi River, China. The developed SCF method can capture discrete and nonlinear relationships between explanatory and response variables. Cluster trees are generated through the SCF method to reflect complex relationships between independent (i.e. explanatory) and dependent (i.e. response) variables in the hydrologic system without determining specific linear/nonlinear functions. The developed SCF method is applied for monthly streamflow prediction in Xiangxi River based on the local meteorological records as well as some climate index. Comparison among SCF, multiple linear regression, generalized regression neural network, and least square support vector machine methods would be conducted. The results indicate that the SCF method would produce good predictions in both training and testing periods. Besides, the inherent probabilistic characteristics of the SCF predictions are further analyzed. The results obtained by SCF can presented as intervals, formulated by the minimum and maximum predictions as well as the 5 and 95 % percentile values of the predictions, which can reflect the variations in streamflow forecasts. Therefore, the developed SCF method can be applied for monthly streamflow prediction in various watersheds with complicated hydrologic processes.     

Estimation of suspended sediment concentration and yield using linear models,random forests and quantile regression forests

For sediment yield estimation, intermittent measurements of suspended sediment concentration (SSC) have to be interpolated to derive a continuous sedigraph. Traditionally, sediment rating curves (SRCs) based on univariate linear regression of discharge and SSC (or the logarithms thereof) are used but alternative approaches (e.g. fuzzy logic, artificial neural networks, etc.) exist. This paper presents a comparison of the applicability of traditional SRCs, generalized linear models (GLMs) and non‐parametric regression using Random Forests (RF) and Quantile Regression Forests (QRF) applied to a dataset of SSC obtained for four subcatchments (0·08, 41, 145 and 445 km²) in the Central Spanish Pyrenees. The observed SSCs are highly variable and range over six orders of magnitude. For these data, traditional SRCs performed inadequately due to the over‐simplification of relating SSC solely to discharge. Instead, the multitude of acting processes required more flexibility to model these nonlinear relationships. Thus, alternative advanced machine learning techniques that have been successfully applied in other disciplines were tested. GLMs provide the option of including other relevant process variables (e.g. rainfall intensities and temporal information) but require the selection of the most appropriate predictors. For the given datasets, the investigated variable selection methods produced inconsistent results. All proposed GLMs showed an inferior performance, whereas RF and QRF proved to be very robust and performed favourably for reproducing sediment dynamics. QRF additionally provides estimates on the accuracy of the predictions and thus allows the assessment of uncertainties in the estimated sediment yield that is not commonly found in other methods. The capabilities of RF and QRF concerning the interpretation of predictor effects are also outlined. Copyright © 2008 John Wiley & Sons, Ltd.     

Potential of support vector regression for prediction of monthly streamflow using endogenous property

In the recent past, a variety of statistical and other modelling approaches have been developed to capture the properties of hydrological time series for their reliable prediction. However, the extent of complexity hinders the applicability of such traditional models in many cases. Kernel‐based machine learning approaches have been found to be more popular due to their inherent advantages over traditional modelling techniques including artificial neural networks(ANNs ). In this paper, a kernel‐based learning approach is investigated for its suitability to capture the monthly variation of streamflow time series. Its performance is compared with that of the traditional approaches. Support vector machines (SVMs) are one such kernel‐based algorithm that has given promising results in hydrology and associated areas. In this paper, the application of SVMs to regression problems, known as support vector regression (SVR), is presented to predict the monthly streamflow of the Mahanadi River in the state of Orissa, India. The results obtained are compared against the results derived from the traditional Box–Jenkins approach. While the correlation coefficient between the observed and predicted streamflows was found to be 0·77 in case of SVR, the same for different auto‐regressive integrated moving average (ARIMA) models ranges between 0·67 and 0·69. The superiority of SVR as compared to traditional Box‐Jenkins approach is also explained through the feature space representation. Copyright © 2009 John Wiley & Sons, Ltd.     

Prediction of flow rate of karstic springs using support vector machines

Complex void space structure and flow patterns in karstic aquifers render behaviour prediction of karstic springs difficult. Four support vector regression-based models are proposed to predict flow rates from two adjacent karstic springs in Greece (Mai Vryssi and Pera Vryssi). Having no accurate estimates of the groundwater flow pattern, we used four kernels: linear, polynomial, Gaussian radial basis function and exponential radial basis function (ERBF). The data used for training and testing included daily and mean monthly precipitation, and spring flow rates. The support vector machine (SVM) performance depends on hyper-parameters, which were optimized using a grid search approach. Model performance was evaluated using root mean square error and correlation coefficient. Polynomial kernel performed better for Mai Vryssi and the ERBF for Pera Vryssi. All models except one performed better for Pera Vryssi. Our models performed better than generalized regression neural network, radial basis function neural network and ARIMA models.     

A non-linear and non-stationary perspective for downscaling mean monthly temperature: a wavelet coupled second order Volterra model

This study presents a multiscale framework for downscaling of the General Circulation Model (GCM) outputs to the mean monthly temperature at regional scale using a wavelet based Second order Voltera (SoV) model. The models are developed using the reanalysis climatic data from the National Centers for Environmental Prediction (NCEP) and are validated using the simulated climatic dataset from the Can CM4 GCM for five locations in the Krishna river basin, India. K-means clustering, based on the multiscale wavelet entropy of the predictors, is used for obtaining the clusters of the input climatic variables. Principal component analysis (PCA) is used to obtain the representative variables from each cluster. These input variables are then used to develop a wavelet based multiscale model using Second order Volterra approach to simulate observed mean monthly temperature for the selected locations in the basin. These models are called W-P-SoV models in this paper. For the purpose of comparison, linear multi-resolution models are developed using Multiple Linear regression (MLR) and are called W-P MLR models. The performance of the models is further compared with other Wavelet-PCA based models coupled with Multiple linear regression models (P-MLR) and Artificial Neural Networks (P-ANN), and, stand-alone MLR and ANN to establish the superiority of the proposed approach. The results indicate that the performance of the wavelet based models is superior in terms of downscaling accuracy when compared with the other models used.     

Extreme learning machine for the displacement prediction of landslide under rainfall and reservoir level

Landslide prediction is always the emphasis of landslide research. Using global positioning system GPS technologies to monitor the superficial displacements of landslide is a very useful and direct method in landslide evolution analysis. In this paper, an EEMD–ELM model [ensemble empirical mode decomposition (EEMD) based extreme learning machine (ELM) ensemble learning paradigm] is proposed to analysis the monitoring data for landslide displacement prediction. The rainfall data and reservoir level fluctuation data are also integrated into the study. The rainfall series, reservoir level fluctuation series and landslide accumulative displacement series are all decomposed into the residual series and a limited number of intrinsic mode functions with different frequencies from high to low using EEMD technique. A novel neural network technique, ELM, is employed to study the interactions of these sub-series at different frequency affecting landslide occurrence. Each sub-series extracted from accumulative displacement of landslide is forecasted respectively by establishing appropriate ELM model. The final prediction result is obtained by summing up the calculated predictive displacement value of each sub. The EEMD–ELM model shows the best accuracy comparing with basic artificial neural network models through forecasting the displacement of Baishuihe landslide in the Three Gorges reservoir area of China.     

Monthly suspended sediment load prediction using artificial intelligence: testing of a new random subspace method

ABSTRACT

The predictive capability of a new artificial intelligence method, random subspace (RS), for the prediction of suspended sediment load in rivers was compared with commonly used methods: random forest (RF) and two support vector machine (SVM) models using a radial basis function kernel (SVM-RBF) and a normalized polynomial kernel (SVM-NPK). Using river discharge, rainfall and river stage data from the Haraz River, Iran, the results revealed: (a) the RS model provided a superior predictive accuracy (NSE = 0.83) to SVM-RBF (NSE = 0.80), SVM-NPK (NSE = 0.78) and RF (NSE = 0.68), corresponding to very good, good, satisfactory and unsatisfactory accuracies in load prediction; (b) the RBF kernel outperformed the NPK kernel; (c) the predictive capability was most sensitive to gamma and epsilon in SVM models, maximum depth of a tree and the number of features in RF models, classifier type, number of trees and subspace size in RS models; and (d) suspended sediment loads were most closely correlated with river discharge (PCC = 0.76). Overall, the results show that RS models have great potential in data poor watersheds, such as that studied here, to produce strong predictions of suspended load based on monthly records of river discharge, rainfall depth and river stage alone.     

Streamflow forecasting using least-squares support vector machines

Abstract

This paper investigates the ability of a least-squares support vector machine (LSSVM) model to improve the accuracy of streamflow forecasting. Cross-validation and grid-search methods are used to automatically determine the LSSVM parameters in the forecasting process. To assess the effectiveness of this model, monthly streamflow records from two stations, Tg Tulang and Tg Rambutan of the Kinta River in Perak, Peninsular Malaysia, were used as case studies. The performance of the LSSVM model is compared with the conventional statistical autoregressive integrated moving average (ARIMA), the artificial neural network (ANN) and support vector machine (SVM) models using various statistical measures. The results of the comparison indicate that the LSSVM model is a useful tool and a promising new method for streamflow forecasting.

Editor D. Koutsoyiannis; Associate editor L. See

Citation Shabri, A. and Suhartono, 2012. Streamflow forecasting using least-squares support vector machines. Hydrological Sciences Journal, 57 (7), 1275–1293.     

Deep echo state network: a novel machine learning approach to model dew point temperature using meteorological variables

ABSTRACT

The potential of different models – deep echo state network (DeepESN), extreme learning machine (ELM), extra tree (ET), and regression tree (RT) – in estimating dew point temperature by using meteorological variables is investigated. The variables consist of daily records of average air temperature, atmospheric pressure, relative humidity, wind speed, solar radiation, and dew point temperature (T_dew) from Seoul and Incheon stations, Republic of Korea. Evaluation of the model performance shows that the models with five and three-input variables yielded better accuracy than the other models in these two stations, respectively. In terms of root-mean-square error, there was significant increase in accuracy when using the DeepESN model compared to the ELM (18%), ET (58%), and RT (64%) models at Seoul station and the ELM (12%), ET (23%), and RT (49%) models at Incheon. The results show that the proposed DeepESN model performed better than the other models in forecasting T_dew values.     

Application of Multiboost-KELM algorithm to alleviate the collinearity of log curves for evaluating the abundance of organic matter in marine mud shale reservoirs: a case study in Sichuan Basin,China

The total organic carbon (TOC) content reflects the abundance of organic matter in marine mud shale reservoirs and reveals the hydrocarbon potential of the reservoir. Traditional TOC calculation methods based on statistical and machine learning have limited effect in improving the computational accuracy of marine mud shale reservoirs. In this study, the collinearity between log curves of marine mud shale reservoirs was revealed for the first time, which was found to be adverse to the improvement of TOC calculation accuracy. To this end, a new TOC prediction method was proposed based on Multiboost-Kernel extreme learning machine (Multiboost-KELM) bridging geostatistics and machine learning technique. The proposed method not only has good data mining ability, generalization ability and sound adaptivity to small samples, but also has the ability to improve the computational accuracy by reducing the effect of collinearity between logging curves. In prediction of two mud shale reservoirs of Sichuan basin with proposed model, the results showed that the predicted value of TOC was in good consistence with the measured value. The root-mean-square error of TOC predicting results was reduced from 0.415 (back-propagation neural networks) to 0.203 and 1.117 (back-propagation neural networks) to 0.357, respectively; the relative error value decreased by up to 8.9%. The Multiboost-KELM algorithm proposed in this paper can effectively improve the prediction accuracy of TOC in marine mud shale reservoir.     

Forecasting monthly precipitation using sequential modelling

In the hydrological cycle, rainfall is a major component and plays a vital role in planning and managing water resources. In this study, new generation deep learning models, recurrent neural network (RNN) and long short-term memory (LSTM), were applied for forecasting monthly rainfall, using long sequential raw data for time series analysis. “All-India” monthly average precipitation data for the period 1871–2016 were taken to build the models and they were tested on different homogeneous regions of India to check their robustness. From the results, it is evident that both the trained models (RNN and LSTM) performed well for different homogeneous regions of India based on the raw data. The study shows that a deep learning network can be applied successfully for time series analysis in the field of hydrology and allied fields to mitigate the risks of climatic extremes.     

Suspended sediment yield estimation using genetic algorithm-based artificial intelligence models: case study of Mahanadi River,India

The estimation of sediment yield is important in design, planning and management of river systems. Unfortunately, its accurate estimation using traditional methods is difficult as it involves various complex processes and variables. This investigation deals with a hybrid approach which comprises genetic algorithm-based artificial intelligence (GA-AI) models for the prediction of sediment yield in the Mahanadi River basin, India. Artificial neural network (ANN) and support vector machine (SVM) models are developed for sediment yield prediction, where all parameters associated with the models are optimized using genetic algorithms simultaneously. Water discharge, rainfall and temperature are used as input to develop the GA-AI models. The performance of the GA-AI models is compared to that of traditional AI models (ANN and SVM), multiple linear regression (MLR) and sediment rating curve (SRC) method for evaluating the predictive capability of the models. The results suggest that GA-AI models exhibit better performance than other models.     

Long-range forecasting of suspended sediment

This study examines the employment of two methods, multiple linear regression (MLR) and an artificial neural network (ANN), for multistep ahead forecasting of suspended sediment. The autoregressive integrated moving average (ARIMA) model is considered for one-step ahead forecasting of sediment series in order to provide a comparison with the MLR and ANN methods. For one- and two-step ahead forecasting, the ANN model performance is superior to that of the MLR model. For longer ranges, MLR models provide better accuracy, but there is an important assumption violation. The Durbin-Watson statistics of the MLR models show a noticeable decrease from 1.3 to 0.5, indicating that the residuals are not dependent over time. The scatterplots of the three methods (MLR, ARIMA and ANN) for one-step ahead forecasting for the validation period illustrate close fits with the regression line, with the ANN configuration having a slightly higher R² value.     

Rapid estimation of soil heavy metal nickel content based on optimized screening of near-infrared spectral bands

In near-infrared spectroscopy,the traditional feature band extraction method has certain limitations.Therefore,a band extraction method named the three-step extraction method was proposed.This method combines characteristic absorption bands and correlation coefficients to select characteristic bands corresponding to various spectral forms and then uses stepwise regression to eliminate meaningless variables.Partial least squares regression(PLSR)and extreme learning machine(ELM)models were used to verify the effect of the band extraction method.Results show that the differential transformation of the spectrum can effectively improve the correlation between the spectrum and nickel(Ni)content.Most correlation coefficients were above 0.7 and approximately 20%higher than those of other transformation methods.The model effect established by the feature variable selection method based on comprehensive spectral transformation is only slightly affected by the spectral transformation form.Infive types of spectral transformation,the RPD values of the proposed method were all within the same level.The RPD values of the PLSR model were concentrated between 1.6 and 1.8,and those of the ELM model were between 2.5 and2.9,indicating that this method is beneficial for extracting more complete spectral features.The combination of the three-step extraction method and ELM algorithm can effectively retain important bands associated with the Ni content of the soil.The model based on the spectral band selected by the three-step extraction method has better prediction ability than the other models.The ELM model of the first-order differential transformation has the best prediction accuracy(RP^2=0.923,RPD=3.634).The research results provide some technical support for monitoring heavy metal content spectrum in local soils.     

Adjusting wavelet‐based multiresolution analysis boundary conditions for long‐term streamflow forecasting

We propose a novel technique for improving a long‐term multi‐step‐ahead streamflow forecast. A model based on wavelet decomposition and a multivariate Bayesian machine learning approach is developed for forecasting the streamflow 3, 6, 9, and 12 months ahead simultaneously. The inputs of the model utilize only the past monthly streamflow records. They are decomposed into components formulated in terms of wavelet multiresolution analysis. It is shown that the model accuracy can be increased by using the wavelet boundary rule introduced in this study. A simulation study is performed to evaluate the effects of different wavelet boundary rules using synthetic and real streamflow data from the Yellowstone River in the Uinta Basin in Utah. The model based on the combination of wavelet and Bayesian machine learning regression techniques is compared with that of the wavelet and artificial neural networks‐based model. The robustness of the models is evaluated. Copyright © 2015 John Wiley & Sons, Ltd.