A comparison of performance of several artificial intelligence methods for forecasting monthly discharge time series

Developing a hydrological forecasting model based on past records is crucial to effective hydropower reservoir management and scheduling. Traditionally, time series analysis and modeling is used for building mathematical models to generate hydrologic records in hydrology and water resources. Artificial intelligence (AI), as a branch of computer science, is capable of analyzing long-series and large-scale hydrological data. In recent years, it is one of front issues to apply AI technology to the hydrological forecasting modeling. In this paper, autoregressive moving-average (ARMA) models, artificial neural networks (ANNs) approaches, adaptive neural-based fuzzy inference system (ANFIS) techniques, genetic programming (GP) models and support vector machine (SVM) method are examined using the long-term observations of monthly river flow discharges. The four quantitative standard statistical performance evaluation measures, the coefficient of correlation (R), Nash–Sutcliffe efficiency coefficient (E), root mean squared error (RMSE), mean absolute percentage error (MAPE), are employed to evaluate the performances of various models developed. Two case study river sites are also provided to illustrate their respective performances. The results indicate that the best performance can be obtained by ANFIS, GP and SVM, in terms of different evaluation criteria during the training and validation phases.     

A hidden Markov model segmentation procedure for hydrological and environmental time series

In this paper we present a procedure for the segmentation of hydrological and enviromental time series. We consider the segmentation problem from a purely computational point of view which involves the minimization of Huberts segmentation cost; in addition this least squares segmentation is equivalent to Maximum Likelihood segmentation. Our segmentation procedure maximizes Likelihood and minimizes Huberts least squares criterion using a hidden Markov model (HMM) segmentation algorithm. This algorithm is guaranteed to achieve a local maximum of the Likelihood. We evaluate the segmentation procedure with numerical experiments which involve artificial, temperature and river discharge time series. In all experiments, the procedure actually achieves the global minimum of the Likelihood; furthermore execution time is only a few seconds, even for time series with over a thousand terms.     

A comparison of three methods for regional flood frequency analysis in Saudi Arabia

The objective of the study was to compare the relative accuracy of three methodologies of regional flood frequency analysis in areas of limited flood records. Thirty two drainage basins of different characteristics, located mainly in the southwest region of Saudi Arabia, were selected for the study. In the first methodology, region curves were developed and used together with the mean annual flood, estimated from the characteristics of drainage basin, to estimate flood flows at a location in the basin. The second methodology was to fit probability distribution functions to annual maximum rainfall intensity in a drainage basin. The best fitted probability function was used together with common peak flow models to estimate the annual maximum flood flows in the basin. In the third methodology, duration reduction curves were developed and used together with the average flood flow in a basin to estimate the peak flood flows in the basin. The results obtained from each methodology were compared to the flood records of the selected stations using three statistical measures of goodness-of-fit. The first methodology was found best in a case of having short length of record at a drainage basin. The second methodology produced satisfactory results. Thus, it is recommended in areas where data are not sufficient and/or reliable to utilise the first methodology.     

Groundwater system response at sites with different agricultural land uses: case of the Guarani Aquifer outcrop area,Brotas/SP-Brazil

A system identification approach can be incorporated in groundwater time series analysis, revealing information concerning the local hydrogeological situation. The aim of this work was to analyse water table fluctuations in an outcrop area of the Guarani Aquifer System (GAS) in Brotas/SP, Brazil, using data from a groundwater monitoring network. The water table dynamic was modelled using continuous time series models that reference the hydrogeological system upon which they operate. The model’s climatological inputs of precipitation and evapotranspiration generate impulse response (IR) functions with parameters that can be related to the physical conditions concerning the hydrological processes involved. The interpretation of the model parameters from two sets of monitoring wells selected at different land-use sites is presented, exemplifying the effect of different water table depths and the distance to the nearest drainage location. Systematic trends of water table depths were also identified from model parameters at specific periods and related to plant development, crop harvest and land-use changes.

EDITOR D. Koutsoyiannis

ASSOCIATE EDITOR L. Ruiz     

Temporal analysis of the frequency and duration of low and high streamflow: years of record needed to characterize streamflow variability

A temporal analysis of the number and duration of exceedences of high- and low-flow thresholds was conducted to determine the number of years required to detect a level shift using data from Virginia, North Carolina, and South Carolina. Two methods were used—ordinary least squares assuming a known error variance and generalized least squares without a known error variance. Using ordinary least squares, the mean number of years required to detect a one standard deviation level shift in measures of low-flow variability was 57.2 (28.6 on either side of the break), compared to 40.0 years for measures of high-flow variability. These means become 57.6 and 41.6 when generalized least squares is used. No significant relations between years and elevation or drainage area were detected (P>0.05). Cluster analysis did not suggest geographic patterns in years related to physiography or major hydrologic regions. Referring to the number of observations required to detect a one standard deviation shift as ‘characterizing’ the variability, it appears that at least 20 years of record on either side of a shift may be necessary to adequately characterize high-flow variability. A longer streamflow record (about 30 years on either side) may be required to characterize low-flow variability.     

Comparative analysis of predictive methods for drought durations: a case of monthly and annual streamflow droughts in Atlantic Canada

Methods based on the recursive probability, the extreme number theorem, and Markov chain (MC) concepts were applied to predict drought lengths (duration) on the standardized (termed as standardized hydrological index, SHI) sequences of monthly and annual river flows from Atlantic Canada. Results of the study indicated that the MC-based method is the most efficient, reliable and versatile method for predicting drought durations followed by the extreme-number-based method. The recursive-probability-based method was found to be computationally intensive and less efficient, although it provided a powerful means for calibrating the empirical plotting position formula needed in the MC-based method. The Weibull plotting position formula turned out to be a suitable measure of the exceedance probability in MC methodology for predicting drought lengths in Atlantic Canada. Based on results, it can be inferred that the MC-based method can be extended to MC2 and higher-order chains for predicting drought lengths on SHI sequences. The predictive capability of the extreme-number-theorem-based method is limited only to independent or weakly first-order persistent SHI sequences.

EDITOR D. Koutsoyiannis

ASSOCIATE EDITOR Q. Zhang     

A comparison of methods to estimate groundwater recharge from bare soil based on data observed by a large-scale lysimeter

Effectively estimating groundwater recharge is critical to manage water resources, especially in arid and semi-arid regions as impacted by intensive human activities and climate changes. Rare insights have been gained into groundwater recharge since direct observation is hard to carry out. Although several methods are currently available to estimate groundwater recharge, the estimated results may cover noticeable bias. The behaviours of different methods based on different conceptual frameworks and exhibiting different levels of complexity should be examined to estimate actual groundwater recharge. This study aims to assess the performance of four common methods to estimate groundwater recharge. For this end, large-scale lysimeters equipped with soil water content sensors and water table sensors were set up at a research site established in Guanzhong Basin of China. The data achieved by 1-year observation were employed to compare four estimation methods. As revealed from the results, the following findings are drawn. (a) Groundwater level fluctuation (GLF) method is simple, whereas its accuracy is determined by specific yield, and adopting a water balance method to estimate specific yield can considerably enhance the accuracy of GLF. (b) The calibrated numerical model can obtain the optimal result compared with the other methods, whereas long-term observation data are required for parameter calibration. (c) In the water balance method, the maximum entropy production (MEP) model and a practical method (estimating evaporation between two rainfall events) were used to calculate evaporation. As indicated by the results, water balance method combined with MEP is capable of obtaining more reliable results of groundwater recharge compared with the practical method. (d) With an analytical model based on linearized Richards' equation, accurate results can be achieved. What is more, the analytical model only needs the measurement of soil moisture near the surface. The limitation of this method is that it is difficult to determine the maximal water flux. The mentioned findings are of critical implications to the management and sustainable development of groundwater.     

A predictive model for Arias intensity at multiple sites and consideration of spatial correlations

Arias intensity, I_a, has been identified as an efficient intensity measure for the estimation of earthquake‐induced losses. In this paper, a new model for the prediction of Arias intensity, which incorporates nonlinear site response through the use of the average shear‐wave velocity and a heteroskedastic variance structure, is proposed. In order to estimate the effects of ground motions on spatially‐distributed systems, it is important to take into account the spatial correlation of the intensity measure. However, existing loss‐estimation models, which use I_a as an input, do not take this aspect of the ground motion into account. Therefore, the potential to model the spatial correlation of Arias intensity is also investigated. The empirical predictive model is developed using recordings from the Pacific Earthquake Engineering Research Center Next Generation of Attenuation database whereas the model for spatial correlation makes use of the well‐recorded events from this database, that is the Northridge and Chi‐Chi earthquakes. Copyright © 2011 John Wiley & Sons, Ltd.     

数值模拟静态应力扰动下的断层失稳:结果分析兼与Dieterich模型和Coulomb模型的对比

基于大量岩石力学实验,Dieterich和其他研究者(Dieterich, 1978; Ruina, 1983)首先提出了描述岩石摩擦过程的速率-状态摩擦定律(R-S摩擦定律).如今R-S摩擦定律已成为研究地震成核等地震演化机制的有效手段.Dieterich (1992, 1994)最早提出了描述受静态剪应力扰动后断层失稳时间提前或推后的余震触发机制的解析模型.现在Dieterich模型已经成为解释余震随时间衰减规律的Omori定律等地震观测现象的有力工具.与之相对应,广泛使用的Coulomb应力失稳模型也可以给出断层受到静态剪应力扰动后,断层失稳时间的提前和推后量.Dieterich模型和Coulomb应力失稳模型基于不同的物理方法,所以在进行地震危险性评估时,二者均有各自的局限性.本文利用R-S摩擦定律控制的1-D弹簧-滑块模型,模拟计算了理论地震循环以及在不同静态剪切应力扰动下,失稳时间的提前和推后量的变化情况,然后将计算得到的时间提前和推后量分别与Dieterich模型和Coulomb模型的相应计算结果进行了定量化对比和差异性分析,并给出了相应的解释.数值模拟结果显示,对于R-S摩擦定律在参数不变的条件下,断层模型失稳时间的提前和推后量的大小强烈依赖于静态剪应力扰动的大小和作用时间,而且绝对值相同的正、负向静态剪应力扰动造成的失稳时间的提前和推后量的变化情况并不完全一致.在震后松弛/滑移阶段和闭锁阶段,时间提前和推后量是常数,且随静态剪应力扰动绝对值的增大而增大,两者的比值接近于1.0,这与Dieterich模型和Coulomb模型的结果是一致的,相应的差值小于两模型结果的10%.而在自加速阶段,模拟计算结果则存在与Dieterich模型和Coulomb模型结果不同的特征.首先,在自加速阶段模拟计算结果均偏离Coulomb模型,而且时间提前和推后量的比值小于1.0,相异于Coulomb模型的论断.不过当受到正向静态剪应力扰动后,Dieterich模型的结果和模拟计算结果是一致的,最大相差量不超过Dieterich模型结果的7%,可接近0.对于负向静态剪应力扰动,当其绝对值较小时,Dieterich模型的结果很接近模拟计算结果,相差量不超过该结果的14%.但对于绝对值较大的情况,模拟计算结果远大于Dieterich模型的结果,最大可达Dieterich模型结果的35倍,这是由于负向静态剪应力扰动后使得Ω=δθ/dc>>1的条件不再成立,进而使得Dieterich模型不再成立.总的来说,与模拟计算相比Dieterich模型可以很好地描述1-D断层受扰后失稳时间提前和推后量的变化情况,并且可以体现出正、负静态剪切应力扰动后失稳时间提前量和推后量变化的差异性,而Coulomb模型则不能完整地给出受到静态剪应力扰动后断层失稳时间提前或推后的估计值.     

电离层预报模型研究

当利用无线电电磁波进行远程通信、卫星导航时,传递信号要受到电离层的影响,因此,对电离层中电子含量的研究显得特别重要.虽然国际上有几种电离层的电子含量预报模型,但其预报只能精确到电子含量的50%～60%.本文提出了一种新的电离层电子含量预报方法:即用球谐函数对IGS(国际GPS服务)所给出的离地面450 km高的球面上的每一网点的电离层电子含量进行拟合,对不同的时间所得到的拟合系数所形成的时间序列用时间序列分析理论中的ARMA(p,q)模型进行预报,从而实现全球的电离层电子含量预报.利用本方法对2004年和2005年IGS所给电离层电子含量资料在地理框架中做了分析预报,5天内电子含量预报相对精度在90%左右.     

Multiwell study of seismic attenuation at the CO2CRC Otway project geosequestration site: Comparison of amplitude decay,centroid frequency shift and 1D waveform inversion methods

At the CO2CRC Otway geosequestration site, the abundance of borehole seismic and logging data provides a unique opportunity to compare techniques of Q (measure of attenuation) estimation and validate their reliability. Specifically, we test conventional time-domain amplitude decay and spectral-domain centroid frequency shift methods versus the 1D waveform inversion constrained by well logs on a set of zero-offset vertical seismic profiles. The amplitude decay and centroid frequency shift methods of Q estimation assume that a seismic pulse propagates in a homogeneous medium and ignore the interference of the propagating wave with short-period multiples. The waveform inversion explicitly models multiple scattering and interference on a stack of thin layers using high-resolution data from sonic and density logs. This allows for stable Q estimation in small depth windows (in this study, 150 m), and separation of the frequency-dependent layer-induced scattering from intrinsic absorption. Besides, the inversion takes into account band-limited nature of seismic data, and thus, it is less dependent on the operating frequency bandwidth than on the other methods. However, all considered methods of Q estimation are unreliable in the intervals where subsurface significantly deviates from 1D geometry. At the Otway site, the attenuation estimates are distorted by sub-vertical faults close to the boreholes. Analysis of repeated vertical seismic profiles reveals that 15 kt injection of the CO₂-rich fluid into a thin saline aquifer at 1.5 km depth does not induce detectable absorption of P-waves at generated frequencies 5–150 Hz, most likely because the CO₂ plume in the monitoring well is thin, <15 m. At the Otway research site, strong attenuation Q ≈ 30–50 is observed only in shaly formations (Skull Creek Mudstone, Belfast Mudstone). Layer-induced scattering attenuation is negligible except for a few intervals, namely 500–650 m from the surface, and near the injection interval, at around 1400–1550 m, where Q_scat ≈ 50–65.     

Identification and classification of transient pulses observed in magnetometer array data by time-domain principal component analysis filtering

A method for identification of pulsations in time series of magnetic field data which are simultaneously present in multiple channels of data at one or more sensor locations is described. Candidate pulsations of interest are first identified in geomagnetic time series by inspection.Time series of these ‘‘training events' ' are represented in matrix form and transpose-multiplied to generate timedomain covariance matrices. The ranked eigenvectors of this matrix are stored as a feature of the pulsation. In the second stage of the algorithm, a sliding window(approximately the width of the training event) is moved across the vector-valued time-series comprising the channels on which the training event was observed. At each window position, the data covariance matrix and associated eigenvectors are calculated. We compare the orientation of the dominant eigenvectors of the training data to those from the windowed data and flag windows where the dominant eigenvectors directions are similar. This was successful in automatically identifying pulses which share polarization and appear to be from the same source process. We apply the method to a case study of continuously sampled(50 Hz) data from six observatories, each equipped with threecomponent induction coil magnetometers. We examine a90-day interval of data associated with a cluster of four observatories located within 50 km of Napa, California,together with two remote reference stations-one 100 km to the north of the cluster and the other 350 km south. When the training data contains signals present in the remote reference observatories, we are reliably able to identify and extract global geomagnetic signals such as solar-generatednoise. When training data contains pulsations only observed in the cluster of local observatories, we identify several types of non-plane wave signals having similar polarization.     

A model of the redistribution of disaggregated soil particles by rainsplash

A model of the dispersion of splash droplets from a single raindrop impact on a sloping soil surface is combined with a theory of the entrainment of mineral particles from a disaggregated mixture in splash droplets to obtain a model of the dispersion of such particles by a raindrop impact. Stochastic modelling techniques extend this further to a model of the spatial redistribution of soil on a plot after a period of rainfall. Since the model is probabilistic and physically based it enables the incorporation of further advances in the understanding of splash erosion at all stages and can simulate the effect of the stochastic nature of rainfall and soil properties on the process. Several different situations are simulated. These include the movement of marked soil particles from point sources and the spatial patterns of erosion on a sloping plot. The model can also simulate the differential erosion of different soil particle size fractions.     

用EOF分析方法研究强震前中国大陆水平位移场时空分布特征

利用大气科学常用的经验正交函数(EOF)分解方法, 对中国大陆连续观测的基准站的水平位移每周解时间序列进行分析。 首先对时间序列中不连续的数据进行内插处理, 并通过线性拟合从时间序列中去掉长期滑动速率的影响。 结合震例研究强震前水平位移场在NS向和EW向的时间和空间的分布特征。     

Conditional resampling of hydrologic time series using multiple predictor variables: A K-nearest neighbour approach

Unlike parametric alternatives for time series generation, non-parametric approaches generate new values by conditionally resampling past observations using a probability rationale. Observations lying ‘close’ to the conditioning vector are resampled with higher probability, ‘closeness’ is defined using a Euclidean or Mahalanobis distance formulation. A common problem with these approaches is the difficulty in distinguishing the importance of each predictor in the estimation of the distance. As a consequence, the conditional probability and hence the resampled series, can offer a biased representation of the true population it aims to simulate. This paper presents a variation of the K-nearest neighbour resampler designed for use with multiple predictor variables. In the modification proposed, an influence weight is assigned to each predictor in the conditioning set with the aim of identifying nearest neighbours that represent the conditional dependence in an improved manner. The workability of the proposed modification is tested using synthetic data from known linear and non-linear models and its applicability is illustrated through an example where daily rainfall is downscaled over 15 stations near Sydney, Australia using a predictor set consisting of selected large-scale atmospheric circulation variables.     

A discrete particle representation of hillslope hydrology: hypothesis testing in reproducing a tracer experiment at Gårdsjön,Sweden

Despite the long history of the continuum equation approach in hydrology, it is not a necessary approach to the formulation of a physically based representation of hillslope hydrology. The Multiple Interacting Pathways (MIPs) model is a discrete realization that allows hillslope response and transport to be simultaneously explored in a way that reflects the potential occurrence of preferential flows and lengths of pathways. The MIPs model uses random particle tracking methods to represent the flow of water within the subsurface alongside velocity distributions that acknowledge preferential flows and transition probability matrices, which control flow pathways. An initial realization of this model is presented here in application to a tracer experiment carried out in Gårdsjön, Sweden. The model is used as an exploratory tool, testing several hypotheses in relation to this experiment. Copyright © 2011 John Wiley & Sons, Ltd.     

A comparison of instrumental dewatering methods for the separation and concentration of suspended sediment for subsequent trace element analysis

A comparison involving both field and laboratory trials was performed to evaluate the utility of two continuous-flow centrifuges and a tangential-flow filtration system for dewatering suspended sediments for subsequent trace element analysis. Although recovery efficiencies for the various devices differ, the analytical results from the separated suspended sediments indicate that any of the tested units can be used effectively and precisely for dewatering. Further, the three devices appear to concentrate and dewater suspended sediments in such a manner as to be equivalent to that which could be obtained by in-line filtration. Only the tangential-flow filtration system appears capable of providing both a dewatered sediment sample and a potentially usable effluent, which can be analysed for dissolved trace elements. The continuous-flow centrifuges can process whole water at an influent feed rate of 41 per minute; however, when suspended sediment concentrations are low (<30mg^?1), when small volumes of whole water are to be processed (30 to 401), or when suspended sediment mean grain size is very fine (<10 μm), influent feed rates of 21 per minute may be more efficient. Tangential-flow filtration can be used to process samples at the rate of 11 per minute.     

Cluster analysis of soil CO<Subscript>2</Subscript> data from Mt. Etna (Italy) reveals volcanic influences on temporal and spatial patterns of degassing

Soil CO₂ concentration data were collected periodically from July 2001 to June 2005 from sampling site grids in two areas located on the lower flanks of Mt. Etna volcano (Paternò and Zafferana Etnea–Santa Venerina). Cluster analysis was performed on the acquired data in order to identify possible groups of sites where soil degassing could be fed by different sources. In both areas three clusters were recognised, whose average CO₂ concentration values throughout the whole study period remained significantly different from one another. The clusters with the lowest CO₂ concentrations showed time-averaged values ranging from 980 to 1,170 ppm vol, whereas those with intermediate CO₂ concentrations showed time-averaged values ranging from 1,400 to 2,320 ppm vol, and those with the highest concentrations showed time-averaged values between 1,960 and 55,430 ppm vol. We attribute the lowest CO₂ concentrations largely to a biogenic source of CO₂. Conversely, the highest CO₂ concentrations are attributed to a magmatic source, whereas the intermediate values are due to a variable mixing of the two sources described above. The spatial distribution of the CO₂ values related to the magmatic source define a clear direction of anomalous degassing in the Zafferana Etnea–Santa Venerina area, which we attribute to the presence of a hidden fault, whereas in the Paternò area no such oriented anomalies were observed, probably because of the lower permeability of local soil. Time-series analysis shows that most of the variations observed in the soil CO₂ data from both areas were related to changes in the volcanic activity of Mt. Etna. Seasonal influences were only observed in the time patterns of the clusters characterised by low CO₂ concentrations, and no significant interdependence was found between soil CO₂ concentrations and meteorological parameters. The largest observed temporal anomalies are interpreted as release of CO₂ from magma batches that migrated from deeper to shallower portions of Etna’s feeder system. The pattern of occurrence of such episodes of anomalous gas release during the observation period was quite different between the two studied areas. This pattern highlighted an evident change in the mechanism of magma transport and storage within the volcano’s feeder system after June 2003, interpreted as magma accumulation into a shallow (<8 km depth) reservoir.     

北方半干旱区土地利用/覆被变化对湖泊水质的影响:以岱海流域为例(2000-2018年)

土地利用/覆被变化对明晰气候变化和人类活动对湖泊水环境的影响有重要作用.以北方典型农牧交错的岱海流域为研究对象,基于遥感解译技术、马尔可夫转移矩阵、综合污染指数法等方法,对2000-2018年岱海流域土地利用/覆被和湖泊水质的变化进行分析,并结合冗余分析法和计量分析模型探究长时间序列尺度下土地利用/覆被变化对湖泊水质的...