Investigating the extrapolation performance of neural network models in suspended sediment data

Suspended sediment modelling is a quite significant issue in hydrology. The prediction of suspended sediment has taken the attention of several scientists in water resources. With extrapolation, the forecasting ability of the employed forecasting model beyond the calibration range is investigated. In the present study, different smoothing parameters are used to differentiate the kurtosis of the local critical points (local minima and maxima). The two models used are an artificial neural network (ANN) model and a multiple linear regression (MLR) model for prediction in order to examine the model extrapolation ability. The ANN model provides closer estimations to the observed peaks, being higher than the corresponding MLR ones. For the local minima, the ANN predictions are higher than the MLR predictions. As there are limited local points, all the remaining ANN predictions are lower than the MLR ones except for one point.     

Application of newly developed ensemble machine learning models for daily suspended sediment load prediction and related uncertainty analysis

ABSTRACT

Ensemble machine learning models have been widely used in hydro-systems modeling as robust prediction tools that combine multiple decision trees. In this study, three newly developed ensemble machine learning models, namely gradient boost regression (GBR), AdaBoost regression (ABR) and random forest regression (RFR) are proposed for prediction of suspended sediment load (SSL), and their prediction performance and related uncertainty are assessed. The SSL of the Mississippi River, which is one of the major world rivers and is significantly affected by sedimentation, is predicted based on daily values of river discharge (Q) and suspended sediment concentration (SSC). Based on performance metrics and visualization, the RFR model shows a slight lead in prediction performance. The uncertainty analysis also indicates that the input variable combination has more impact on the obtained predictions than the model structure selection.     

A modelling approach for estimating suspended sediment concentrations for multiple rivers influenced by agriculture

A data-driven model based on an adaptive neuro-fuzzy inference system (ANFIS) was tested for the estimation of suspended sediment concentrations within watersheds influenced by agriculture. ANFIS models were developed using different combinations of inputs such as precipitation, streamflow, surface runoff and the watershed vulnerability index. A multi-watershed ANFIS model was also developed combining the datasets from all studied watersheds. The best results were obtained from a combination of precipitation, streamflow and watershed vulnerability index as input variables. Nash-Sutcliffe coefficients were improved for the multi-watershed ANFIS compared to watershed-specific ANFIS models. The introduction of the erosion vulnerability index significantly improved the ability of the ANFIS model to estimate suspended sediment concentrations within the watersheds. Furthermore, the inclusion of this index opens the possibility of using the ANFIS model to investigate the impact of land-use changes on sediment delivery.     

An evaluation of sediment rating curves for estimating suspended sediment concentrations for subsequent flux calculations

In the absence of actual suspended sediment concentration (SSC) measurements, hydrologists have used sediment rating (sediment transport) curves to estimate (predict) SSCs for subsequent flux calculations. Various evaluations of the sediment rating‐curve method were made using data from long‐term, daily sediment‐measuring sites within large (>1 000 000 km²), medium (<1 000 000 to >1000 km²), and small (<1000 km²) river basins in the USA and Europe relative to the estimation of suspended sediment fluxes. The evaluations address such issues as the accuracy of flux estimations for various levels of temporal resolution as well as the impact of sampling frequency on the magnitude of flux estimation errors. The sediment rating‐curve method tends to underpredict high, and overpredict low SSCs. As such, the range of errors associated with concomitant flux estimates for relatively short time‐frames (e.g. daily, weekly) are likely to be substantially larger than those associated with longer time‐frames (e.g. quarterly, annually) because the over‐ and underpredictions do not have sufficient time to balance each other. Hence, when error limits must be kept under ±20%, temporal resolution probably should be limited to quarterly or greater. The evaluations indicate that over periods of 20 or more years, errors of <1% can be achieved using a single sediment rating curve based on data spanning the entire period. However, somewhat better estimates for the entire period, and markedly better annual estimates within the period, can be obtained if individual annual sediment rating curves are used instead. Relatively accurate (errors <±20%) annual suspended sediment fluxes can be obtained from hydrologically based monthly measurements/samples. For 5‐year periods or longer, similar results can be obtained from measurements/samples collected once every 2 months. In either case, hydrologically based sampling, as opposed to calendar‐based sampling is likely to limit the magnitude of flux estimation errors. Published in 2003 John Wiley & Sons, Ltd.     

Real-time prediction of the peak suspended sediment concentration and sediment yield of the Lao-Nung River during storms

Based on rainfall erosion of soil and suspended sediment transport in storm events, a method is proposed to predict peak suspended sediment concentration and suspended sediment yield in watersheds based on rainfall characteristics prior to peak rainfall intensity. The rainfall characteristics factors that dominate peak suspended sediment concentration C_p are rainfall erosion factor R_ef, first peak rainfall intensity of area-average rainfall i_p1 and antecedent precipitation index I_ap; the rainfall characteristics factors that dominate suspended sediment yield Y_ss in storm events are total rainfall P, suspended sediment yield factor R_sf and antecedent precipitation index I_ap. This research focuses on watersheds in Liau-Kwei observation station along Lao-Nung River in southern Taiwan as the research object, and adopts the PSED-model to simulate the discharge hydrograph, suspended sediment concentration hydrograph and suspended sediment yield in 11 storm events for analysis. The analytical results show that there is a good correlation between the above-mentioned rainfall characteristics factors and C_p as well as Y_ss, thus enabling C_p and Y_ss to be predicted by using Expressions (13) and (14). These two expressions are utilized to predict C_p and Y_ss of Typhoon Morakot in 2009, and the results are compared with those from simulation by using the PSED-model. The result of comparison shows there is a good capability in predicting. For the watersheds where it is necessary to predict C_p and Y_ss of a storm event for the benefit of effective operation of water resource facilities, the aforesaid rainfall characteristics factors can be utilized to establish applicable models for prediction.     

Analyses of suspended sediment loads in Slovenian rivers

ABSTRACT

Suspended solids are present in every river, but high quantities can worsen the ecological conditions of streams; therefore, effective monitoring and analysis of this hydrological variable are necessary. Frequency, seasonality, inter-correlation, extreme events, trends and lag analyses were carried out for peaks of suspended sediment concentration (SSC) and discharge (Q) data from Slovenian streams using officially monitored data from 1955 to 2006 that were made available by the Slovenian Environment Agency. In total more than 500 station-years of daily Q and SSC data were used. No uniform (positive or negative) trend was found in the SSC series; however, all the statistically significant trends were decreasing. No generalization is possible for the best fit distribution function. A seasonality analysis showed that most of the SSC peaks occurred in the summer (short-term intense convective precipitation produced by thunderstorms) and in the autumn (prolonged frontal precipitation). Correlations between Q and SSC values were generally relatively small (Pearson correlation coefficient values from 0.05 to 0.59), which means that the often applied Q–SSC curves should be used with caution when estimating annual suspended sediment loads. On average, flood peak Q occurred after the corresponding SSC peak (clockwise-positive hysteresis loops), but the average lag time was rather small (less than 1 day).

Editor M.C. Acreman; Associate editor Y. Gyasi-Agyei     

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.     

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.     

Capacitive sensing technique for silt suspended sediment concentration monitoring

Automated, real-time, and continuous techniques for monitoring suspended sediment concentration in rivers and reservoirs can play an important role in the improvement of the quantity and quality of sediment data, and are valuable to the management of water environment, water conservancy, hazard prevention, and water resources. Research in the monitoring techniques has examined the possibility of using the characteristics of dielectric constants for detecting soil moisture and concentration of air-water two-phase flow, based on the fact that dielectric constants of sediment, air and water are different. A capacitance sensor was developed to monitor the silt suspended sediment concentration (SSSC) in a recent study, following the principle that as SSSC increases in the sediment-water mixture, the apparent dielectric constant of the water sample also increases and therefore the capacitance detected by the sensing system also increases. It is demonstrated that the variations in the concentration of silt sediment correlates positively with the variations in observed capacitance in a linear fashion, and correlates negatively with voltage outputs but also in a linear fashion. The correlation coefficients reached above 0.98. The overall errors in estimated concentrations range between 0.26% and 2.91%. Elements in the capacitance sensor system such as the frequencies of the signal generating system, areas of the electrode plates, and effects of sample temperature have also been evaluated. The results illustrated that the capacitance sensor techniques can be applied to monitoring SSSC automatically and continuously. Also, the range of SSSC in the experiment reached 200 kg/m³; therefore, the application of this technique in practical SSSC monitoring is worthy of further research.     

Evaluation of a simple,inexpensive, in situ sampler for measuring time‐weighted average concentrations of suspended sediment in rivers and streams

The accurate measurement of suspended sediment (<200 μm) in aquatic environments is essential to understand and effectively manage changes to sediment, nutrient, and contaminant concentrations on both temporal and spatial scales. Commonly used sampling techniques for suspended sediment either lack the ability to accurately measure sediment concentration (e.g., passive sediment samplers) or are too expensive to deploy in sufficient number to provide landscape‐scale information (e.g., automated discrete samplers). Here, we evaluate a time‐integrated suspended sediment sampling technique, the pumped active suspended sediment (PASS) sampler, which collects a sample that can be used for the accurate measurement of time‐weighted average (TWA) suspended sediment concentration and sediment particle size distribution. The sampler was evaluated against an established passive time‐integrated suspended sediment sampling technique (i.e., Phillips sampler) and the standard discrete sampling method (i.e., manual discrete sampling). The PASS sampler collected a sample representative of TWA suspended sediment concentration and particle size distribution of a control sediment under laboratory conditions. Field application of the PASS sampler showed that it collected a representative TWA suspended sediment concentration and particle size distribution during high flow events in an urban stream. The particle size distribution of sediment collected by the PASS and Phillips samplers were comparable and the TWA suspended sediment concentration of the samples collected using the PASS and discrete sampling techniques agreed well, differing by only 4% and 6% for two different high flow events. We should note that the current configuration of the PASS sampler does not provide a flow‐weighted measurement and, therefore, is not suitable for the determination of sediment loads. The PASS sampler is a simple, inexpensive, and robust in situ sampling technique for the accurate measurement of TWA suspended sediment concentration and particle size distribution.     

Modelling suspended sediment concentration using artificial neural networks for Gangotri glacier

The dynamics of suspended sediment involves inherent non‐linearity and complexity because of existence of both spatial variability of the basin characteristics and temporal climatic patterns. This complexity, therefore, leads to inaccurate prediction by the conventional sediment rating curve (SRC) and other empirical methods. Over past few decades, artificial neural networks (ANNs) have emerged as one of the advanced modelling techniques capable of addressing inherent non‐linearity in the hydrological processes. In the present study, feed‐forward back propagation (FFBP) algorithm of ANNs is used to model stage–discharge–suspended sediment relationship for ablation season (May–September) for melt runoff released from Gangotri glacier, one of the largest glaciers in Himalaya. The simulations have been carried out on primary data of suspended sediment concentration (SSC) discharge and stage for ablation season of 11‐year period (1999–2009). Combinations of different input vectors (viz. stage, discharge and SSC) for present and previous days are considered for development of the ANN models and examining the effects of input vectors. Further, based on model performance indices for training and testing phase, a suitable modelling approach with appropriate model input structure is suggested. The conventional SRC method is also used for modelling discharge–sediment relationship and performance of developed models is evaluated by statistical indices, namely; root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R²). Statistically, the performance of ANN‐based models is found to be superior as compared to SRC method in terms of the selected performance indices in simulating the daily SSC. The study reveals suitability of ANN approach for simulation and estimation of daily SSC in glacier melt runoff and, therefore, opens new avenues of research for application of hybrid soft computing models in glacier hydrology. Copyright © 2015 John Wiley & Sons, Ltd.     

近岸Ⅱ类水体表层悬浮泥沙浓度遥感模式研究进展

因为具有明显的时间与空间分辨率优势,遥感数据成为近岸Ⅱ类水体悬浮泥沙浓度(SSC)信息提取研究的重要数据源之一.悬浮泥沙遥感信息提取的现状可归纳为:(1)建立近岸Ⅱ类水体SSC遥感模式的方法有三种类型,分别是基于地面光谱与SSC测量的反射率反演方法、基于图像信息法和基于大气辐射传输理论模型法;(2)基于地面测量的反射率反演方法属于理论与经验相结合的方法,也是目前用于SSC定量化遥感模式研究的常用方法.其数学表达形式包括线性关系式、对数关系式、负指数关系式、Gordon模式和综合模式等;(3)到目前为止已有的Ⅱ类水体SSC遥感模式适用性方面还不理想,远未达到与试验室分析相匹配的精度.文章认为:加强地面水文光谱实验研究,建立多光谱SSC定量模式,以高分辨率和高光谱遥感融合数据为基础的SSC定量遥感是今后该方向发展趋势.     

The influence of climate change on suspended sediment transport in Danish rivers

The HIRHAM regional climate model suggests an increase in temperature in Denmark of about 3 °C and an increase in mean annual precipitation of 6–7%, with a larger increase during winter and a decrease during summer between a control period 1961–1990 and scenario period 2071–2100. This change of climate will affect the suspended sediment transport in rivers, directly through erosion processes and increased river discharges and indirectly through changes in land use and land cover. Climate‐change‐induced changes in suspended sediment transport are modelled for five scenarios on the basis of modelled changes in land use/land cover for two Danish river catchments: the alluvial River Ansager and the non‐alluvial River Odense. Mean annual suspended sediment transport is modelled to increase by 17% in the alluvial river and by 27% in the non‐alluvial for steady‐state scenarios. Increases by about 9% in the alluvial river and 24% in the non‐alluvial river were determined for scenarios incorporating a prolonged growing season for catchment vegetation. Shortening of the growing season is found to have little influence on mean annual sediment transport. Mean monthly changes in suspended sediment transport between ? 26% and + 68% are found for comparable suspended sediment transport scenarios between the control and the scenario periods. The suspended sediment transport increases during winter months as a result of the increase in river discharge caused by the increase in precipitation, and decreases during summer and early autumn months. Copyright © 2007 John Wiley & Sons, Ltd.     

Artificial neural network simulation for prediction of suspended sediment concentration in the River Ramganga,Ganges Basin,India

The relation between the water discharge (Q) and suspended sediment concentration (SSC) of the River Ramganga at Bareilly, Uttar Pradesh, in the Himalayas, has been modeled using Artificial Neural Networks (ANNs). The current study validates the practical capability and usefulness of this tool for simulating complex nonlinear, real world, river system processes in the Himalayan scenario. The modeling approach is based on the time series data collected from January to December (2008-2010) for Q and SSC. Three ANNs (T1-T3) with different network configurations have been developed and trained using the Levenberg Marquardt Back Propagation Algorithm in the Matlab routines. Networks were optimized using the enumeration technique, and, finally, the best network is used to predict the SSC values for the year 2011. The values thus obtained through the ANN model are compared with the observed values of SSC. The coefficient of determination (R2), for the optimal network was found to be 0.99. The study not only provides insight into ANN modeling in the Himalayan river scenario, but it also focuses on the importance of understanding a river basin and the factors that affect the SSC, before attempting to model it. Despite the temporal variations in the study area, it is possible to model and successfully predict the SSC values with very simplistic ANN models.     

Application of soft-computing techniques in forecasting sediment load and concentration

ABSTRACT

This paper investigates conventional and soft-computing methods for the estimation of suspended sediment concentration (SSC) and load (SSL) in rivers. Frequently used methods of sediment rate curve (SRC) and multi-nonlinear regression, and soft-computing methods of multi-layer perceptron, multi-linear regression and adaptive neuro-fuzzy inference system are implemented using various hydrological and hydraulic parameters for the Little Kickapoo Creek Watershed, Illinois, USA. All methods performed equally well in the estimation of SSL, without any noticeable outperformance from any from the methods. However, the application of soft-computing methods decreased SSC estimation errors considerably as compared to the results of SRC. The results are significant in the way they reconcile traditionally used hydrological parameters into the soft-computing methods. Overall, soft-computing methods are recommended for the estimation of SSC in rivers because of their reasonably better performance and ease of implementation.     

Using ADV backscatter strength for measuring suspended cohesive sediment concentration

Laboratory experiments were conducted at two institutes to reveal the relationship between acoustic backscatter strength and suspended sediment concentration (SSC). In total, three acoustic Doppler velocimeters (ADVs) with different frequencies (5, 10 and 16 MHz) were tested. Two different commercial clays and one natural sediment from Clay Bank site in the York River were checked for acoustic responses. The SSCs of selected sediments were artificially changed between a selected low and a high value in tap or de-ion water. Each ADV showed quite different backscatter responses depending on the sediment type and SSC. Not all devices had a good linear relationship between backscatter strength and SSC. Within a limited range of SSC, however, the backscatter strength can be well correlated with the SSC. Compared with optical backscattering sensor (OBS), the fluctuation of ADV backscatter signals was too noisy to be directly converted to the instantaneous changes of SSC due to high amplification ratio and small sampling volume. For the more accurate signal conversion for finding the fluctuation of SSC, the ensemble average should be applied to increase the signal-to-noise ratio. There are unexpected responses for the averaged backscatter wave strength: (1) high signals from small particles but low signals from large particles; and (2) two linear segments in calibration slope. These phenomena would be most likely caused by the different gain setting built in ADVs. The different acoustic responses to flocculation might also contribute somewhat if flocs are tightly packed. This study suggests that an ADV could be a useful instrument to estimate suspended cohesive sediment concentration and its fluctuation if the above concerns are clarified.     

Physics‐based numerical modelling of large braided rivers dominated by suspended sediment

Physics‐based models have been increasingly developed in recent years and applied to simulate the braiding process and evolution of channel units in braided rivers. However, limited attention is given to lowland braided rivers where the transport of suspended sediment plays a dominant role. In the present study, a numerical model based on the basic physics laws of hydrodynamics and sediment transport is used to simulate the evolution process of a braided river dominated by suspended load transport. The model employs a fractional method to simulate the transport of graded sediments and uses a multiple‐bed‐layer approach to represent the sediment sorting process. An idealized braided river has been produced, with the hydrodynamic, sediment transport and morphological processes being analysed. In particular, the formation process of local pool–bar units in the predicted river has been investigated. A sensitivity analysis has also been undertaken to investigate the effects of grid resolution and an upstream perturbation on the model prediction. A variety of methods are applied to analyse the geometrical and topographical properties of the modelled river. Self‐organizing characteristics related to river geometry and topography are analysed by state‐space plots, which indicate a close relationship with the periodical erosion and deposition cycles of braiding. Cross‐sectional topography and slope frequency display similar geometries to natural rivers. Scaling characteristics are found by correlation analysis of bar parameters. Copyright © 2014 John Wiley & Sons, Ltd.     

分季节的太湖悬浮物遥感估测模型研究

根据1996-2002年无锡太湖监测站的水质资料分析,太湖悬浮物具有季节性特征,因而分季节的悬浮物估测模型比单一的模型可能更加适合用来估测太湖全年的悬浮物浓度.在分析太湖水体光谱特征的基础上,根据太湖悬浮物的季节性分布特征,使用春夏秋冬四季的Landsat TM/ETM图像和准同步的水质采样数据,建立了太湖分季节的悬浮物估算模型.结果表明:估测因子(B2 B3)/(B2/B3)在春、秋、冬三季都能很好地估测出悬浮物的浓度(R2＞0.52).夏季由于叶绿素的干扰性较大,悬浮物的估测效果不理想.冬季的估测效果最好(R2=0.81),模型为lnSS=14.656×(B2 B3)/(B2/B3) 1.661,其中,ln SS表示悬浮物取自然对数后的值,B2、B3为TM/ETM图像经过6S大气校正、3×3低通滤波后第2、3波段的反射率值.     

考虑采砂影响的鄱阳湖丰水期悬浮泥沙浓度模拟

针对受采砂活动影响显著的鄱阳湖高浑浊水体,结合数值模拟和遥感技术,利用已有的鄱阳湖采砂区遥感监测结果,在构建的鄱阳湖水动力-悬浮泥沙输移模型中添加泥沙点源,对2011年7月1-31日采砂影响下的鄱阳湖丰水期悬浮泥沙浓度进行数值模拟.利用悬浮泥沙浓度实测数据和MODIS影像反演结果对模拟结果的有效验证表明,考虑采砂影响后,悬浮泥沙浓度模拟值与实测值具有强相关关系,确定性系数为0.831,均方根误差为15.5 mg/L,悬浮泥沙浓度空间分布趋势与遥感反演结果基本一致.模拟结果显示,采砂活动对鄱阳湖南部主湖区、河流入湖口影响较小,其主要影响由南向北,经棠荫以西和松门山岛以北航道、入江水道延伸到湖口区域,是鄱阳湖北湖区高浑浊水体形成的重要原因.     

Estimation of suspended sediment concentration in the Saint John River using rating curves and a machine learning approach

ABSTRACT

Sedimentation in navigable waterways and harbours is of concern for many water and port managers. One potential source of variability in sedimentation is the annual sediment load of the river that empties in the harbour. The main objective of this study was to use some of the regularly monitored hydro-meteorological variables to compare estimates of hourly suspended sediment concentration in the Saint John River using a sediment rating curve and a model tree (M5?) with different combinations of predictors. Estimated suspended sediment concentrations were multiplied by measured flows to estimate suspended sediment loads. Best results were obtained using M5? with four predictors, returning an R² of 0.72 on calibration data and an R² of 0.46 on validation data. Total load was underestimated by 1.41% for the calibration period and overestimated by 2.38% for the validation period. Overall, the model tree approach is suggested for its relative ease of implementation and constant performance.

EDITOR M.C. Acreman; ASSOCIATE EDITOR B. Touaibia