Long short-term memory for predicting daily suspended sediment concentration

Frequent and accurate estimation of suspended sediment concentration (SSC) in surface waters and hydraulic schemes is of prime importance for proper design, operation and management of many hydraulic projects. in the present study, a long short-term memory (LSTM) was considered for predicting daily suspended sediment concentration in a river. The LSTM extends recurrent neural network with memory cells, instead of recurrent units, to store and output information, easing the learning of temporal relationships on long time scales. To build the model, daily observed time series of river discharge (Q) and SSC in the Schuylkill River in the United States were used. The results of the proposed model were evaluated and compared with the feedforward neural network and the adaptive neuro fuzzy inference system models which were trained using three different learning algorithms and widely used in the literature for prediction of daily SSC. The comparison of prediction accuracy of the models demonstrated that the LSTM model could satisfactory predict SSC time series, and adequately estimate cumulative suspended sediment load (SSL).

     

Remote sensing retrieval of suspended sediment concentration in shallow waters

The dynamics of coastal lagoons and estuarine areas is characterized by a delicate balance between biological and physical processes and the comprehension and monitoring of such processes require observations over a wide range of temporal and spatial scales. Remote sensing techniques in this context are very advantageous and potentially allow overcoming the spatial limitations of traditional in situ point observations, providing new opportunities for a better understanding of the relevant bio-geomorphological processes and for the calibration and validation of spatially-distributed hydrodynamic and transport models. Remote sensing of suspended particulate matter (SPM) concentration in shallow waters must, however, overcome the difficulties associated with i) the influence of bottom reflection, which may interfere with an accurate retrieval; ii) the necessity of accurately knowing the optical properties of the suspended matter, and iii) the importance of providing an assessment of the uncertainty associated with the estimates produced. This work presents a method to estimate SPM concentration in lagoon/estuarine waters by use of a simplified radiative transfer model. We use a calibration/validation method based on cross-validation and bootstrap techniques to provide a statistically sound determination of model parameters and an evaluation of the uncertainty induced by their inaccurate determination as well as by the uncertain knowledge of the bottom sediment reflectance. The method is applied to the Venice lagoon, using observations from a network of turbidity sensors and from several multispectral satellite sensors (LANDSAT, ASTER and ALOS AVNIR). The bootstrap and cross-validation procedures employed show that consistent estimates of SPM concentration can indeed be retrieved from satellite remote sensing, provided that sufficient in situ ancillary information for appropriate calibration is available. The quantification of the estimation uncertainty shows that retrievals obtained from remote sensing are accurate, robust and repeatable. The SPM concentration maps produced show a general coherence with known features in the Venice lagoon and, together with suitable biological information, point to the role played by benthic vegetation in the stabilization of the bottom sediment.     

Spectral irradiance profiles of suspended marine clay for the estimation of suspended sediment concentration in tropical waters

Reflectance spectra of water containing varying concentrations of organic and inorganic sediments isolated from the coastal waters of Singapore were measured using a portable spectroradiometer under controlled experimental conditions and natural sunlight. The effect of different sizes of sediments on the spectral profile of water was also investigated. In the presence of organic sediments, maximum reflectance of water was observed over a broad band between 440-580 nm, with a peak reflectance at about 550 nm, followed by two smaller peaks at 600-645 nm and 665-690 nm. In contrast, inorganic sediments produced a distinct band peak between 595 and 690 nm. For both sediment types, characteristic features in the infra-red region include a reflectance trough at 754 nm and peak at 814 nm. The empirical colour ratios, OD550-OD754 and OD 595-OD754, were used to estimate the organic and inorganic sediment concentrations respectively. In the case of organic sediments, a power function was found to fit the data well ( R 2 =0.89-0.98), whereas a linear fit was found for inorganic sediments ( R 2 =0.53-0.86). In general, decreases in particle size resulted in overall increases in spectral reflectance.     

Quantifying suspended sediment concentration in subglacial sediment plumes discharging from two Svalbard tidewater glaciers using Landsat-8 and in situ measurements

Marine-terminating outlet glaciers discharge mass through iceberg calving, submarine melting, and meltwater run-off. While calving can be quantified by in situ and remote-sensing observations, meltwater run-off, the subglacial transport of meltwater, and submarine melting are not well constrained due to inherent difficulties observing the subglacial and proglacial environments at tidewater glaciers. Remote-sensing and in situ measurements of surface sediment plumes, and their suspended sediment concentration (SSC), have been used as a proxy for glacier meltwater run-off. However, this relationship between satellite reflectance and SSC has predominantly been established using land-terminating glaciers. Here, we use two Svalbard tidewater glaciers to establish a well-constrained relationship between Landsat-8 surface reflecance and SSC and argue that it can be used to measure relative meltwater run-off at tidewater glaciers throughout a summer melt season. We find the highest correlation between SSCs and Landsat-8 surface reflectance by using the red + NIR band combination (r² = 0.76). The highest correlation between SSCs and in situ field spectrometer measurements is in the 740–800 nm wavelength range (r² = 0.85), a spectral range not currently measured by Landsat. Additionally, we find that in situ and Landsat-8 measurements for surface reflectance of SSCs are not interchangeable and therefore establish a relationship for each detection method. We then use the Landsat-8 relationship to calculate total surface sediment load, finding a strong correlation between total surface sediment load and a proxy for meltwater run-off (r² ≥ 0.89). Our results establish a new metric to calculate SSCs from Landsat-8 surface reflectance and demonstrate how the SSC of subglacial sediment plumes can be used to monitor relative seasonal meltwater discharge at tidewater glaciers.     

Estimating suspended sediment concentrations from spectral reflectance data

Abstract

A method for estimating sediment concentrations at different depths in coastal waters from measured spectral reflectance above the sea is proposed. The water backscattering coefficient is obtained by integrating a scattering cross section derived from Mie theory over a Junge type particle-size distribution. The results for turbid water, near an arid coast, with vertical mixing, shows a maximum of particle concentration in the middle layer, while the same method applied to reflectance data for ‘blue’ water, gives a relative concentration of sediment that gradually increases with depth.     

An optimal model for estimating suspended sediment concentration from Landsat TM images in the Caofeidian coastal waters

Suspended sediment concentration (SSC) is one of the most critical parameters in water quality and environmental evaluations. Remote sensing has the potential for monitoring the dynamics and spatial distribution of SSC efficiently. The primary objective of this study is to develop retrieval models that are reliable and sensitive to SSC levels in the Caofeidian area, a new seaport in northeast China, based on Landsat-5 Thematic Mapper (TM) images and a set of in situ data sets, including spectral reflectance data and water quality data. The study finds that the band reflectance ratio and binary combination factor (i.e. the ratio of the reflectance to the particle size) are more effective than single band reflectance, and a non-linear model is more potent than a linear model for predicting SSC in the Caofeidian waters. A quadratic polynomial regression model of the R_TM3/R_TM2 ratio is proposed as the optimal retrieval model after evaluating various models with respect to different sensitive factors. The accuracy of the model is acceptable with a relative error and a root mean square error of 25.35% and 7.22 mg l^–1, respectively; the correlation coefficient between the observed and estimated SSCs is 0.986. This study also indicates that the band reflectance ratio and binary combination factor are effective in weakening and even partially eliminating the effects of the changes in the sediment type (i.e. particle size and refractive index). And the band reflectance ratio is more efficient. Using the proposed model and TM data, SSC levels for the entire region were estimated. Such results can serve as a baseline for future environmental monitoring efforts.     

The spectral responses of algal chlorophyll in water with varying levels of suspended sediment

The purpose of this paper is to investigate the spectral responses of algal chlorophyll and water, under natural sunlight with varying suspended sediment concentrations (SSC). Twenty levels of SSC with each of two sediment types were generated, ranging from 50 to 1000 mgl^?1, in 75101 of water containing chlorophyll-a concentrations of 718 μgl^?1 and 295 μgl^?1. Results indicate that suspended sediments do not eliminate the prominent spectral patterns of algal chlorophyll, even as SSC reached 1000 mgl^?1. Between 400 and 900 nm, the relation between reflectance and SSC satisfies the expression: d²R(λ)/dS²<0. The effects of varying SSC on the positions and magnitudes of pronounced chlorophyll features were investigated. The ratio between the NIR and red wavelengths was totally independent of SSC. Thus, our finding supports using it as an index for measuring chlorophyll in natural surface water containing suspended sediments.     

Toward universal multispectral suspended sediment algorithms

A data acquisition and analysis program has been undertaken to demonstrate the feasibility of remote multispectral techniques for monitoring suspended sediment concentrations in natural water bodies. Two hundred surface radiance measurements (400–1000 nm) were made at Lake Mead with coincident water sampling for laboratory analysis. Water volume spectral reflectance is calculated from the recorded surface radiance and volume reflectance-suspended sediment relationships investigated. Statistical analysis indicates that quantitative estimates of nonfilterable residue and nephelometric turbidity can be obtained from volume spectral reflectance data with sufficient accuracy (based on U.S. Environmental Protection Agency standards) to make the multispectral technique feasible for sediment monitoring. Algorithms exhibit sufficient universality to indicate they can be implemented in many cases with little or no ground truth for calibration.     

加权交叉验证神经网络在水质预测中的应用

在之前的研究中使用人工神经网络进行水质指标预测已经取得一定效果,在此基础上将交叉验证应用于人工神经网络的训练,获得更加准确的预测结果。以澧水某监测站的水质实测数据作为样本,选取总磷、总氮、溶解氧等6个指标,建立水质预测模型。在运用Levenberg-Marquardt优化算法对学习样本进行优化的基础上,采用加权的k-fold交叉验证方法来构建神经网络集合,构建集合时采取三种不同的混合方式：平均值、中间值和加权累积。针对不同的指标,进行了一系列的实验,总的来说,新的预测方法与简单0倍验证相比有更好的预测结果,在所有指标中氨氮和溶解氧含量预测准确率比其他指标高。     

Amazon River time series of surface sediment concentration from MODIS

Moderate Resolution Imaging Spectroradiometer (MODIS) data are of sufficient quality – especially the 8-day composite surface reflectance products – that 19–21 images are consistently available per year for derivation of surface sediment concentration (SSC) for the Amazon River and its tributaries. MODIS-derived SSC from 2001 show similar patterns to field measurements of depth-integrated suspended sediment, with the lowest values occurring during peak flows, and the highest during rising water periods. Tributary data show more irregular patterns, but are consistent among the two datasets. When compared to the few field measurements of SSC, the MODIS data are of similar range and accuracy (7–130 ±10 mg L^?1) at 17 sites in the main channel, large tributaries and floodplain lakes. Combined with remote measures of discharge or stage, this framework could be used to regularly monitor suspended sediment loads on large rivers.     

Adaptive neuro-fuzzy computing technique for suspended sediment estimation

This paper investigates the accuracy of an adaptive neuro-fuzzy computing technique in suspended sediment estimation. The monthly streamflow and suspended sediment data from two stations, Kuylus and Salur Koprusu, in Kizilirmak Basin in Turkey are used as case studies. The estimation results obtained by using the neuro-fuzzy technique are tested and compared with those of the artificial neural networks and sediment rating curves. Root mean squared errors, mean absolute errors and correlation coefficient statistics are used as comparing criteria for the evaluation of the models’ performances. The comparison results reveal that the neuro-fuzzy models can be employed successfully in monthly suspended sediment estimation.     

Testing of a model for the simulation of the volume reflectance of water due to suspended sediment under controlled conditions,for various sediment types

Abstract

A model that simulates the volume reflectance of water is tested in laboratory conditions with three different sediments of known grain size distribution and spectral reflectance. The model uses, as its main input variables, the grain size distribution of the suspended sediment, the density of the grains and the spectral reflectance of the wet sediment from which the suspension comes. The spectral bands of the model are completely programmable and they have been adapted to correspond to green, red and near-infrared, as in Landsat MSS bands 4, 5 and 6. The grain size distribution, the reflectance in the three spectral bands and the density of the three sediment types had been measured and these measurements were used as input variables for the model to simulate the volume reflectance of the water, in the same spectral bands, for different suspended sediment concentrations. Each sediment type was used to produce several concentration levels in a specially prepared 1101 tank and the volume reflectance of the water was measured. The measurements were compared with the simulated results of the model. Then the model was modified to simulate the volume reflectance of a stratified environment, where the concentration changes with depth, and used to explain some discrepancies in the chromaticity transformation.     

Prediction of sediment,particulate nutrient and dissolved nutrient concentrations in a dry tropical river to provide input to a mechanistic coastal water quality model

A Generalised Additive Modelling (GAM) approach is applied to prediction of both particulate and dissolved nutrient concentrations in a wet-tropical river (the Fitzroy River, Queensland, Australia). In addition to covariant terms considered in previous work (i.e. flow, discounted flow and a rising-falling limb term), we considered several new potential covariates: meteorological and hydrological variables that are routinely monitored, available in near-real time, and were considered to have potential predictive power. Of the additional terms considered, only flows from three tributaries of the Fitzroy River (namely, the Nogoa, Comet and Isaac Rivers) were found to significantly improve the model. Inclusion of one or more of these additional flow terms greatly improved results for dissolved nitrogen and dissolved phosphorus concentrations, which were not otherwise amenable to prediction. In particular, the Nogoa sub-catchment, dominated by pasture for cattle, was found to be important in determining dissolved inorganic nitrogen and phosphorus concentrations reaching the river mouth. This insight may direct further research, including future refinement of processed-based catchment models. The GAMs described here are used to provide near real-time river boundary conditions for a complex coupled hydrodynamic and biogeochemical model of the Great Barrier Reef Lagoon, and can be coupled with a forecasting hydrological model to allow integrated forecasting simulations of the catchment to coast system.     

Error assessment of validation techniques for estimating suspended particulate matter concentration from airborne multispectral imagery

Significant errors have been reported when transporting algorithms for converting aerial imagery to suspended particulate matter concentration to data collected at different times in different geographical sites. This raises the possibility that the relationships developed are coincidental rather than causal. Determination of the portability of the relationships will allow an assessment of the degree of causality of such algorithms. This paper describes the development of algorithms for two sites in UK coastal waters and the assessment of the errors incurred when porting these algorithms to datasets collected at different seasons and from varying geographical sites. The errors incurred when using algorithms at sites having differing morphologies are considered to be too large for operational use. The high accuracy of transfer to data from the same site at a different season indicates the causality of the relationship derived and allows the recommendation of a way forward for algorithm development.     

Combining deterministic modelling with artificial neural networks for suspended sediment estimates

Estimates of suspended sediment concentrations and transport are an important part of any marine environment assessment study because these factors have a direct impact on the life cycle and survival of marine ecosystems. This paper proposes to implement a combined methodology to tackle these estimates. The first component of the methodology comprised two numerical current and wave models, while the second component was based on the artificial intelligence technique of neural networks (ANNs) used to reproduce values of sediment concentrations observed at two sites. The ANNs were fed with modelled currents and waves and trained to produce area-specific concentration estimates. The trained ANNs were then applied to predict sediment concentrations over an independent period of observations. The use of a data set that merged together observations from both the mentioned sites provided the best ANN testing results in terms of both the normalised root mean square error (0.13) and the mean relative error (0.02).     

A FORTRAN program for suspended sediment dynamics and tidal flux monitoring

Research into estuarine and coastal fine sediment transport and sediment deposition is aided by a proposed routine for collection and analysis of hydrodynamic, particulate, and salinity data. Field measurements taken over a number of depth intervals and over one or more tidal cycles are used to calculate the point, depth-integrated, and net horizontal movements of water, suspended solids, and salt at a fixed sampling site.     

Atmospheric correction of LANDSAT MSS data for a multidate suspended sediment algorithm

This paper describes the application of various processing methods lo a synoptic set of satellite-sea data obtained from the Solent area on the south coast of England. The methods include ‘darkest pixel’ correction, sun angle and radiometric corrections, chromaticity analysis, atmospheric and surface corrections and the use of the satellite sensor response curves. It is shown that the use of a simple atmospheric/surface correction algorithm, based on atmospheric optical transmission theory, provides the most accurate method of estimating estuarine suspended sediment concentration from the satellite data above.     

Identification of suspended sediment in coastal waters using airborne thematic mapper data

Remotely sensed data were collected using an airborne Daedalus thematic mapper, flown at 4000 m, over Swansea Bay, northern Bristol Channel, U.K. The area is one of high tidal range and wave activity. Fifty-eight surface-water/sediment samples were collected from three vessels as concurrent ‘sea-truth’ data. Samples were analysed for suspended sediment concentration (SSC)

Regression equations are developed for SSC using data in the 450-520 nm (channel 2), 520-600 nm (channel 3), 605-625 nm (channel 4) and 630-690 nm (channel 5) spectral bands

Localized flow patterns are visible in the imagery and are identified using suspended material as the passive tracer. Such patterns cannot be identified in temporally averaged data collected from research vessels

It is concluded that airborne remote sensing can contribute to the understanding of the movement of water and sediments on continental shelves, when used in association with conventional oceanographic sampling techniques.     

Retrieval of total suspended matter concentration in the Yellow and East China Seas from MODIS imagery

A retrieval algorithm, of total suspended matter (TSM) concentration in the Yellow Sea (YS) and East China Sea (ECS) was developed using observations made in the 2003 Spring and Autumn cruises over the YS and the ECS. Analysis of the in-situ backscattering coefficients of the suspended particles (b_bp) indicates that the accuracy becomes worse when the concentration of TSM (C_TSM) is higher than 20 mg/l. The accuracy of the b_bp is improved by using a bio-optical model in which b_bp is optimized with a non-linear least-square Levenberg-Marquardt method. The remote sensing reflectance (R_rs) is obtained by means of the optimization. The optimized R_rs for waters with C_TSM higher than 20 mg/l, together with the measured R_rs for waters with C_TSM lower than 20 mg/l, are used to establish the relationships between R_rs(748), R_rs(869) and R_rs(645), which are used in the iterative method for atmospheric correction. Two atmospheric correction algorithms are switched according to the water turbidity. The shortwave infrared wavelengths (SWIR) method is used for waters with high-turbidity, and the iterative method is used otherwise. Results of the atmospheric correction were then applied to the Tassan model modified in this paper to compute the C_TSM. Comparison between the retrieval results from MODIS imagery and the in-situ measurements indicates that the algorithms described in this paper can provide a reliable estimation of the C_TSM distributions in the YS and ECS.