Comparison of local and global approximators in multivariate chaotic forecasting of daily streamflow

ABSTRACT

Although it is conceptually assumed that global models are relatively ineffective in modelling the highly unstable structure of chaotic hydrologic dynamics, there is not a detailed study of comparing the performances of local and global models in a hydrological context, especially with new emerging machine learning models. In this study, the performance of a local model (k-nearest neighbour, k-nn) and, as global models, several recent machine learning models – artificial neural network (ANN), least square-support vector regression (LS-SVR), random forest (RF), M5 model tree (M5), multivariate adaptive regression splines (MARS) – was analysed in multivariate chaotic forecasting of streamflow. The models were developed for Australia’s largest river, the River Murray. The results indicate that the k-nn model was more successful than the global models in capturing the streamflow dynamics. Furthermore, coupled with the multivariate phase-space, it was shown that the global models can be successfully used for obtaining reliable uncertainty estimates for streamflow.     

Verification of ARMA identification for modelling temporal correlations of GNSS observations using the ARMASA toolbox

The classical least-squares (LS) algorithm is widely applied in practice of processing observations from Global Satellite Navigation Systems (GNSS). However, this approach provides reliable estimates of unknown parameters and realistic accuracy measures only if both the functional and stochastic models are appropriately specified. One essential deficiency of the stochastic model implemented in many available GNSS software products consists in neglecting temporal correlations of GNSS observations. Analysing time series of observation residuals resulting from the LS evaluation, the temporal correlation behaviour of GNSS measurements can be efficiently described by means of socalled autoregressive moving average (ARMA) processes. For a given noise realisation, a well-fitting ARMA model can be automatically estimated and identified using the ARMASA toolbox available free of charge in MATLAB® Central.In the preliminary stage of applying the ARMASA toolbox to residual-based modelling of temporal correlations of GNSS observations, this paper presents an empirical performance analysis of the automatic ARMA estimation tool using a large amount of simulated noise time series with representative temporal correlation properties comparable to the GNSS residuals. The results show that the rate of unbiased model estimates increases with data length and decreases with model complexity. For large samples, more than 80% of the identified ARMA models are unbiased. Additionally, the model error representing the deviation between the true data-generating process and the model estimate converges rapidly to the associated asymptotical value for a sufficiently large sample size with respect to the correlation length.     

Dynamics of hourly sea level at Hillarys Boat Harbour,Western Australia: a chaos theory perspective

Water level forecasting using recorded time series can provide a local modelling capability to facilitate local proactive management practices. To this end, hourly sea water level time series are investigated. The records collected at the Hillarys Boat Harbour, Western Australia, are investigated over the period of 2000 and 2002. Two modelling techniques are employed: low-dimensional dynamic model, known as the deterministic chaos theory, and genetic programming, GP. The phase space, which describes the evolution of the behaviour of a nonlinear system in time, was reconstructed using the delay-embedding theorem suggested by Takens. The presence of chaotic signals in the data was identified by the phase space reconstruction and correlation dimension methods, and also the predictability into the future was calculated by the largest Lyapunov exponent to be 437 h or 18 days into the future. The intercomparison of results of the local prediction and GP models shows that for this site-specific dataset, the local prediction model has a slight edge over GP. However, rather than recommending one technique over another, the paper promotes a pluralistic modelling culture, whereby different techniques should be tested to gain a specific insight from each of the models. This would enable a consensus to be drawn from a set of results rather than ignoring the individual insights provided by each model.     

Long‐term river meandering as a part of chaotic dynamics? A contribution from mathematical modelling

In the present contribution we focus our attention on the possible signatures of a chaotic behaviour or a self‐organized criticality state triggered in river meandering dynamics by repeated occurrence of cutoff processes. The analysis is carried out examining, through some robust nonlinear methodologies inferred from time series analysis, both the spatial series of local curvatures and the time series of long‐term channel sinuosity. Temporal distribution of cutoff inter‐arrivals is also investigated. The analyzed data have been obtained by using a suitable physics‐based simulation model for river meandering able to reproduce reasonably the features of real rivers. The results are consistent and show that, at least from a modelling point of view, no evidence of chaotic determinism or self‐organized criticality is detectable in the investigated meandering dynamics. Copyright © 2010 John Wiley & Sons, Ltd.     

The influence of seasonal signals on the estimation of the tectonic motion in short continuous GPS time-series

Most GPS time-series exhibit a seasonal signal that can have an amplitude of a few millimetres. This seasonal signal can be removed by fitting an extra sinusoidal signal with a period of one year to the GPS data during the estimation of the linear trend.However, Blewitt and Lavallée (2002) showed that including an annual signal in the estimation process still can give a larger linear trend error than the trend error estimated from data from which the annual signal has been removed by other means. They assumed that the GPS data only contained white noise and we extend their result to the case of power-law plus white noise which is known to exist in most GPS observations. For the GPS stations CASC, LAGO, PDEL and TETN the difference in trend error between having or not having an annual signal in the data is around ten times larger when a power-law plus white noise model is used instead of a pure white noise model. Next, our methodology can be used to estimate for any station how much the accuracy of the linear trend will improve when one tries to subtract the annual signal from the GPS time-series by using a physical model.Finally, we demonstrate that for short time-series the trend error is more influenced by the fact that the noise properties also need to be estimated from the data. This causes on average an underestimation of the trend error.     

Statistical corrections of spatially interpolated missing precipitation data estimates

Spatial interpolation methods used for estimation of missing precipitation data generally under and overestimate the high and low extremes, respectively. This is a major limitation that plagues all spatial interpolation methods as observations from different sites are used in local or global variants of these methods for estimation of missing data. This study proposes bias‐correction methods similar to those used in climate change studies for correcting missing precipitation estimates provided by an optimal spatial interpolation method. The methods are applied to post‐interpolation estimates using quantile mapping, a variant of equi‐distant quantile matching and a new optimal single best estimator (SBE) scheme. The SBE is developed using a mixed‐integer nonlinear programming formulation. K‐fold cross validation of estimation and correction methods is carried out using 15 rain gauges in a temperate climatic region of the U.S. Exhaustive evaluation of bias‐corrected estimates is carried out using several statistical, error, performance and skill score measures. The differences among the bias‐correction methods, the effectiveness of the methods and their limitations are examined. The bias‐correction method based on a variant of equi‐distant quantile matching is recommended. Post‐interpolation bias corrections have preserved the site‐specific summary statistics with minor changes in the magnitudes of error and performance measures. The changes were found to be statistically insignificant based on parametric and nonparametric hypothesis tests. The correction methods provided improved skill scores with minimal changes in magnitudes of several extreme precipitation indices. The bias corrections of estimated data also brought site‐specific serial autocorrelations at different lags and transition states (dry‐to‐dry, dry‐to‐wet, wet‐to‐wet and wet‐to‐dry) close to those from the observed series. Bias corrections of missing data estimates provide better serially complete precipitation time series useful for climate change and variability studies in comparison to uncorrected filled data series. Copyright © 2013 John Wiley & Sons, Ltd.     

A study on chaotic behaviour of equatorial/low latitude ionosphere over Indian subcontinent,using GPS-TEC time series

The deterministic chaotic behaviour of ionosphere, over Indian subcontinent falling under equatorial/low latitude region, ?0.3 to 22.19°N (geomagnetic), was studied using GPS-TEC time series. The values of Lyapunov exponent are low at Thiruvananthapuram and Agatti (?0.30 and 2.38°N, geomagnetic, respectively), and thereafter increase through Bangalore and Hyderabad (4.14 and 8.54°N, geomagnetic, respectively), and attain maximum at Mumbai (10.09°N, geomagnetic), which is near/at the edge of an anomaly crest. The values of correlation dimension computed for TEC time series are in the range 3.1–3.6, which indicate that equatorial/low latitude ionosphere can be described with four variables. Entropy values estimated for TEC time series show no appreciable latitudinal variabilites. The values of non-linear prediction error exhibit a trough, around the latitude sector, 4.14–16.15°N (Geomagnetic). Based on the values of the above quantifiers, the features of chaotic behaviour of equatorial/low latitude ionosphere are briefly discussed.     

Understanding the time‐varying importance of different uncertainty sources in hydrological modelling using global sensitivity analysis

Simulations from hydrological models are affected by potentially large uncertainties stemming from various sources, including model parameters and observational uncertainty in the input/output data. Understanding the relative importance of such sources of uncertainty is essential to support model calibration, validation and diagnostic evaluation and to prioritize efforts for uncertainty reduction. It can also support the identification of ‘disinformative data’ whose values are the consequence of measurement errors or inadequate observations. Sensitivity analysis (SA) provides the theoretical framework and the numerical tools to quantify the relative contribution of different sources of uncertainty to the variability of the model outputs. In traditional applications of global SA (GSA), model outputs are aggregations of the full set of a simulated variable. For example, many GSA applications use a performance metric (e.g. the root mean squared error) as model output that aggregates the distances of a simulated time series to available observations. This aggregation of propagated uncertainties prior to GSA may lead to a significant loss of information and may cover up local behaviour that could be of great interest. Time‐varying sensitivity analysis (TVSA), where the aggregation and SA are repeated at different time steps, is a viable option to reduce this loss of information. In this work, we use TVSA to address two questions: (1) Can we distinguish between the relative importance of parameter uncertainty versus data uncertainty in time? (2) Do these influences change in catchments with different characteristics? To our knowledge, the results present one of the first quantitative investigations on the relative importance of parameter and data uncertainty across time. We find that the approach is capable of separating influential periods across data and parameter uncertainties, while also highlighting significant differences between the catchments analysed. Copyright © 2016 The Authors. Hydrological Processes. Published by John Wiley & Sons Ltd.     

汶川地震前后四川区域GPS时序特征演变及统计分析

为探索地震事件对GPS坐标时间序列的长周期影响,对汶川地震前后四川GPS观测网络长约10年的解算成果进行了多参数模型噪声特征分析.基于最大似然估计方法和频谱特性分析,提取了地震前后各测站坐标序列中的噪声分量,使用Λ-统计检验,得出"白噪声+闪烁噪声"模型可以作为四川GPS区域观测网络的最优噪声组合模型,同时,地震事件使得地震前后GPS噪声分量中的白噪声、闪烁噪声、随机游走噪声等发生显著改变,说明传统谱噪声分析中简单地将地震数据拼接在一起并进行统一处理的模式并不可取;使用共模误差分析方法、区域速度场变化趋势等信息对地震前后噪声模型的改变成因进行了初步的物理解析.     

Comparative study of monthly inflow prediction methods for the Three Gorges Reservoir

Due to the complexity of influencing factors and the limitation of existing scientific knowledge, current monthly inflow prediction accuracy is unable to meet the requirements of various water users yet. A flow time series is usually considered as a combination of quasi-periodic signals contaminated by noise, so prediction accuracy can be improved by data preprocess. Singular spectrum analysis (SSA), as an efficient preprocessing method, is used to decompose the original inflow series into filtered series and noises. Current application of SSA only selects filtered series as model input without considering noises. This paper attempts to prove that noise may contain hydrological information and it cannot be ignored, a new method that considerers both filtered and noises series is proposed. Support vector machine (SVM), genetic programming (GP), and seasonal autoregressive (SAR) are chosen as the prediction models. Four criteria are selected to evaluate the prediction model performance: Nash–Sutcliffe efficiency, Water Balance efficiency, relative error of annual average maximum (RE_max) monthly flow and relative error of annual average minimum (RE_min) monthly flow. The monthly inflow data of Three Gorges Reservoir is analyzed as a case study. Main results are as following: (1) coupling with the SSA, the performance of the SVM and GP models experience a significant increase in predicting the inflow series. However, there is no significant positive change in the performance of SAR (1) models. (2) After considering noises, both modified SSA-SVM and modified SSA-GP models perform better than SSA-SVM and SSA-GP models. Results of this study indicated that the data preprocess method SSA can significantly improve prediction precision of SVM and GP models, and also proved that noises series still contains some information and has an important influence on model performance.     

Vibration‐based damage assessment of steel structure using global and local response measurements

Damage assessment of a structure involves acquiring and identifying dynamic characteristics of the structure and using these characteristics to evaluate behavior and performance. In this study, an unsymmetrical three‐story steel structure (fabricated with one weak column in the first floor) was tested on shaking table and subjected to a series of earthquake excitations with increasing level of excitation back to back. Besides, white noise excitation was also applied in between the earthquake excitation to serve as the reference state. Both the traditional sensing system (accelerometer and linear variable differential transformer) and the local optical tracker system were implemented in the structure to collect the vibration‐based responses. For operational modal analysis, structural response from white noise excitation will be used in this study. First, the traditional system identification using global response data is used (multivariate autoregressive (AR)‐model) to extract system natural frequencies and mode shapes from all different set of white noise responses after earthquake excitation. The migration of AR‐coefficient ellipse error from each sensor response was used to identify the damage location. Second, blind source separation technique was used to identify the modal contribution of the structure from each test, which provide information to detect the damage severity. Finally, from the local optical tracker array data, the principal component analysis was applied to quantify the earthquake‐induce local stress of the structural member. Combine the result from damage detection using global measurement and the identified local element stress, one can locate and quantify the damage. Copyright © 2015 John Wiley & Sons, Ltd.     

Application of nonlinear forecasting techniques for meteorological modeling

A nonlinear forecasting method was used to predict the behavior of a cloud coverage time series several hours in advance. The method is based on the reconstruction of a chaotic strange attractor using four years of cloud absorption data obtained from half-hourly Meteosat infrared images from Northwestern Spain. An exhaustive nonlinear analysis of the time series was carried out to reconstruct the phase space of the underlying chaotic attractor. The forecast values are used by a non-hydrostatic meteorological model ARPS for daily weather prediction and their results compared with surface temperature measurements from a meteorological station and a vertical sounding. The effect of noise in the time series is analyzed in terms of the prediction results.     

Mitigating parameter bias in hydrological modelling due to uncertainty in covariates

The uncertainty in hydrological model covariates, if ignored, introduces systematic bias in the parameters estimated. We introduce here a method to determine the true value of parameters given uncertainty in model inputs. This method, known as simulation extrapolation (SIMEX) operates on the basis of an empirical relationship between parameters and the level of input noise (or uncertainty). The method starts by generating a series of alternate model inputs by artificially adding white noise in increasing multiples of the known error variance. The resulting parameter sets allow us to formulate an empirical relationship between their values and the level of noise present. SIMEX is based on theory that the trend in alternate parameters can be extrapolated back to the notional error free zone.

We illustrate the strength of SIMEX in improving skills of predictive models that use uncertain sea surface temperature anomaly (SSTA) data over the NINO3 region as predictor to the southern oscillation index (SOI), an alternate measure of the strength of the El Nino southern oscillation. Our hypothesis is that the higher magnitude of noise in the pre 1960 data period introduces bias to model parameters where SSTA is the input variable. The relatively error invariant southern oscillation index (SOI) is regressed over SSTA and calibrated using a subset of the series from 1900 to 1960. We validate the resulting models using the less erroneous 1960–2003 data period. Overall the application of SIMEX is found to reduce the residual predictive errors during the validation period.     

地震报道中死亡人数估计方法的适用性分析

基于国内外8次地震报道死亡人数的统计数据,采用修正指数曲线、龚铂茨曲线、罗吉斯蒂曲线分别进行震后死亡人数估计。对比分析表明三种模型均适合地震报道死亡人数估计;用剔除前两个拟合值后的拟合精度作为预测误差在实际应用中具有实际参考价值。对中国3次强震报道死亡人数拟合结果分析显示,联合三种模型,采用相对误差较小的模型为主、另两种模型为参考的方法进行最终死亡人数预测,可为抗震救灾指挥部署提供参考依据。     

Enhancing gradient-based parameter estimation with an evolutionary approach

Traditionally, the calibration of groundwater models has depended on gradient-based local optimization methods. These methods provide a reasonable degree of success only when the objective function is smooth, second-order differentiable, and satisfies the Lipschitz's condition. For complicated and highly nonlinear objective functions it is almost impractical to satisfy these conditions simultaneously. Research in the calibration of conceptual rainfall-runoff models, has shown that global optimization methods are more successful in locating the global optimum in the region of multiple local optima. In this study, a global optimization technique, known as shuffle complex evolution (SCE), is coupled to the gradient-based Lavenberg–Marquardt algorithm (GBLM). The resultant hybrid global optimization algorithm (SCEGB) is then deployed in parallel testing with SCE and GBLM to solve several inverse problems where parameters of a nonlinear numerical groundwater flow model are estimated. Using perfect (i.e. noise-free) observation data, it is shown SCEGB and SCE are successful at identifying the global optimum and predicting all model parameters; whereas, the commonly applied GBLM fails to identify the optimum. In subsequent inverse simulations using observation data corrupted with noise, SCEGB and SCE again outperform GBLM by consistently producing more accurate parameter estimates. Finally, in all simulations the hybrid SCEGB is seen to be equally effective as SCE but computationally more efficient.     

Neural network simulation of spring flow in karst environments

Daily discharges of two springs lying in a karstic environment were simulated for a period of 2.5 years with the use of a multi-layer perceptron back-propagation neural network. Two models were developed for the springs, one relying on the original data and another where the missing discharge values were supplemented by assuming linear relationships during base flow conditions. For both springs the mean square error of the two models did not differ significantly, with an improvement exhibited at the extremes, during the network’s training phase, by the model that utilized the extended data set, the results of which are reported here. The time lag between precipitation and spring discharge differed significantly for the two springs indicating that in karstic environments hydraulic behavior is dominated, even within a few hundred meters, by local conditions. Optimum training results were attained with a Levenberg–Marquardt algorithm resulting in a network architecture consisting of two input layer neurons, four hidden layer neurons, and one output layer neuron, the spring’s discharge. The neural network’s predictions captured the behavior for both springs and followed very closely the discontinuities in the discharge time series. Under-/over-estimation of observed discharges for the two springs remained below 3 %, with the exception of a few local maxima where the predicted discharges diverged more strongly from observed values. Inclusion of temperature data did not add to the improvement of predictions. Finally, optimum predictions were attained when past discharge data were added to the input record and discharge differentials rather than direct discharges were calculated resulting in elimination of any local maximum discrepancy between observed and predicted discharge values.     

Group-based estimation of missing hydrological data: II. Application to streamflows

Abstract

The group approach that treats hydrological data as groups rather than as single-valued observations was proposed in a companion paper. Various models representing four techniques are briefly presented and applied to single series and bi-series cases, respectively, in this paper. The techniques represented by these models are regression, time series analysis, partitioning modelling, and artificial neural networks. The utility of the models for estimating missing streamflow data using the group approach is investigated. It turns out that the group approach is valid for estimating missing values, and possibly other applications, when data are significantly auto-correlated.     

Spatial and temporal spectra of the geomagnetic field and their scaling properties

Many natural phenomena show a relationship between their spatial and temporal Fourier spectra. This paper discusses such a connection for the geomagnetic field, when some assumptions are made about the (exponential or power-law) behaviour of the spatial power spectrum of the field itself and that of its time derivative (the spatial spectrum of the secular variation) as estimated from global geomagnetic field models. It is shown that, under either assumption, the temporal spectrum of the geomagnetic field computed at the core–mantle boundary (CMB) would have a power-law behaviour with a negative spectral exponent of about 0.5. At the Earth’s surface, although the temporal spectrum obtained from the power-law spatial model assumes a slightly more complicated form, it can be practically approximated with a power law with a negative exponent of about 3.6. Analysis of magnetic observatory data confirms these results and that the starting hypotheses are reasonable, especially in view of the possibly chaotic state of the dynamical processes underlying the generation and maintenance of the geomagnetic field.     

APPLICATION AND STATISTICAL ANALYSIS OF RELATIONSHIP BETWEEN SHEAR WAVE VELOCITY AND DEPTH OF SOIL-LAYERS

The shear wave velocity is one of the important parameters in seismic engineering.The common mathematical models of relationship between shear wave velocity and depth of soil-layers are linear function model,quadratic function model,power function model,cubic function model,and quartic function model.It is generally believed that the regression formulae based on aforementioned mathematical models are mainly used for preliminary estimation of the local shear wave velocity.In order to increase the value of test data of wave speed in boreholes,the calculation formulae for the thickness of ground cover layer are derived based on the aforementioned mathematical models and their fitting parameters.The calculation formulae for the mean shear wave velocity of soil-layers are derived by integral mean value theorem.Accordingly,the calculation formulae for the equivalent shear wave velocity of soil-layers are derived.The calculation formulae for the depth of reflective waves in time-depth conversion of the reflection seismic exploration are derived.Through the statistical analysis of test data of shear wave velocity of soil layers in Changyuan County,Henan Province,regression formulae and their fitting parameters of aforementioned mathematical models are obtained.The results show that in the determination of the quality of these regression formulae and their fitting parameters,the adjusted R-square,root mean square error and residual error,the matching on the statistical range between the geometry of function of mathematical models used and the scattergram of the measured data,the application purpose and the simplicity of the regression formulae should be considered.With the aforementioned new formulae,the results show that the calculated values of equivalent shear wave velocity of soil-layers and thickness of ground cover layer meet the engineering needs.The steps for statistics and applications of the relationship between shear wave velocity and depth of soil-layers for a new area are as follows:(1) Analyze the relevant data about the site such as the drilling and wave speed test data,etc.and divide the site into seismic engineering geological units;(2) In a single seismic engineering geological unit,make statistical analysis of the data of borehole wave speed test,comprehensively identify and select mathematical models and their fitting parameters of the relationship between shear wave velocity and depth of soil-layers;(3) Substitute the selected fitting parameters into the formulae,based on their mathematical models for the thickness of ground cover layer,or the equivalent shear wave velocity of soil-layers,or the depth of reflective wave,then the thickness of ground covering layer,equivalent shear wave velocity,and depth of reflective wave are obtained.