Water quality analysis based on LSTM and BP optimization with a transfer learning model.

In the urban water environmental management, a fast and effective method for water quality analysis should be established with the rapid urbanization. In this study, the Beijing's sub-center was chosen as a case study, and long short-term memory (LSTM) and back propagation (BP) models were built, then a transfer learning model was proposed and applied to optimize the two models on the base of the upstream and downstream relationships in the rivers. The results indicated that the proposed deep learning model could improve NSE by 7% and 9% for LSTM and BP at the Dongguan Bridge gauge, respectively. At the Xugezhuang gauge in the Liangshui River, NSE was improved by 11% and 17%, respectively. At the Yulinzhuang gauge, it was improved by 16% and 13%, respectively. Because the upstream and downstream relationships were considered in the learning model, the model performance was obviously better. In brief, this method would provide an idea for the effective water quality model construction in the ungauged basins or regions.

Evaluation of multiple satellite precipitation products and their potential utilities in the Yarlung Zangbo River Basin.

Hydrological modeling in the Third Pole remains challenging due to the complex topography and scarcity of in-situ precipitation observations. In this study, we assessed five satellite precipitation products (SPPs) including TRMM3B42, PERSIANN-CDR, GPM-IMERG, CMORPH, and GSMaP, and simulated daily streamflow in the Yarlung Zangbo River Basin (YZRB) with VIC model. The performance of SPPs was evaluated by CC, RB, RMSE, POD and FAR, to compare with daily observations. Overall, all SPPs showed decreasing trends of precipitation from east to west compared to 10 km rainfall data. PERSIANN had the highest values of POD (0.65), RB (91.6%) and FAR (0.59) but worst performed in streamflow. CMORPH, GPM and TRMM fit well with the observations annually but overestimate the precipitation in the southeast during wet seasons. Simulation from GPM and CMORPH yield satisfactory results (NSE of 0.86 and 0.82, RE of - 20% and - 13%, respectively), while TRMM outperformed GPM in modeling runoff with smaller relative error. Results indicated the potential of GPM and CMORPH in providing alternative rainfall information in YZRB. Accurate evaluation of multi-source SPPs and their hydrological utility in YZRB would benefit further hydrometeorological studies and water resources management in this area.

Temperature lapse rate estimation and snowmelt runoff simulation in a high-altitude basin.

As a key parameter of hydrological process modeling, the near-surface air temperature lapse rate reflects the vertical changes in air temperature characteristics in alpine basins but often lacks the support of sufficient ground observation data. This study estimated the lapse rate of the Lhasa River Basin (LRB) from the monthly air temperature dataset (2001-2015), which was derived based on good relationships between the observed air temperature at eight gauged stations and the corresponding gridded land surface temperature of MODIS. The estimated annual average air temperature lapse rate was approximately 0.62 °C/100 m. The monthly lapse rate in different years varied seasonally in the range of 0.45-0.8 °C/100 m; the maximum was in May, and the relatively low value occurred from September to January. The snow cover in the zones with relatively low altitudes showed seasonal variation, which was consistent with the air temperature variation. Permanent snow cover appeared in the area above 5000 m and expanded with increasing elevation.

Daily precipitation dataset at 0.1° for the Yarlung Zangbo River basin from 2001 to 2015.

In order to obtain higher precision regional precipitation dataset in the Yarlung Zangbo River basin, two different schemes were proposed on the basis of the two most application potential satellite-based precipitation products, IMERG and CMORPH_BLD. The first method aimed to correct the positive error of IMERG based on high correlation (CC > 0.9) between IMERG and gauges. The second algorithm was developed to merge IMERG with CMORPH_BLD by the stepwise linear regression. As the reference, IMERG played a key role in correction of precipitation ratio determination and precipitation event detection. Two daily datasets with 0.1° resolution (BRD_IMERG and IGREA_IMERG-CMORPH) performed better than IMERG in CC, RMSE, ME, FAR and CSI, and streamflow simulation in the whole basin (NS: 0.86 and 0.87; RBIAS: -19% and -11%) and sub-basins. The two proposed methods were relatively simple and efficient for reconstructing higher precision regional precipitation, and the datasets provided a good application demonstration in the alpine region.

A data-driven framework for spatiotemporal characteristics, complexity dynamics, and environmental risk evaluation of river water quality.

To evaluate the evolution of river water quality in a changing environment, measuring the objective water quality is critical for understanding the rules of river water pollution. Based on the sample entropy theory and a nonlinear statistical method, this study aims to identify the spatiotemporal dynamics of water quality and its complexity in the Yangtze River basin using time series data, to separate the contributions of human activity and climate change to water quality, and to establish a data-driven risk assessment framework for the spatial (potential risk) and temporal (direct risk) aspects of water pollution. The results demonstrate that the spatiotemporal dynamics of water quality and sample entropy in each monitoring section are closely related to the characteristics of the corresponding location. The water quality of the main stream is superior, and its complexity is less than that of the tributaries. Cascade reservoir operation and vegetation status, agricultural production, and rainfall patterns exert great influences in the upper, middle, and lower reaches, respectively. Dam construction, urban agglomeration development, and interactions between river and lake are also influencing factors. An attributional analysis found that climate change and human activities negatively contributed to the evolution of NH3-N concentration in most of the monitored sections, and the average relative contribution rates of human activities to changes in water quality in the main and tributary streams were -55.46% and -48.49%, respectively. In addition, the construction of data-driven risk assessment framework can efficiently and accurately assess the potential and direct water pollution risks of rivers.

Evaluating Different Methods for Determining the Velocity-Dip Position over the Entire Cross Section and at the Centerline of a Rectangular Open Channel.

The velocity profile of an open channel is an important research topic in the context of open channel hydraulics; in particular, the velocity-dip position has drawn the attention of hydraulic scientists. In this study, analytical expressions for the velocity-dip position over the entire cross section and at the centerline of a rectangular open channel are derived by adopting probability methods based on the Tsallis and general index entropy theories. Two kinds of derived entropy-based expressions have the same mathematical form as a function of the lateral distance from the sidewall of the channel or of the aspect ratio of the channel. Furthermore, for the velocity-dip position over the entire cross section of the rectangular open channel, the derived expressions are compared with each other, as well as with two existing deterministic models and the existing Shannon entropy-based expression, using fifteen experimental datasets from the literature. An error analysis shows that the model of Yang et al. and the Tsallis entropy-based expression predict the lateral distribution of the velocity-dip position better than the other proposed models. For the velocity-dip position at the centerline of the rectangular open channel, six existing conventional models, the derived Tsallis and general index entropy-based expressions, and the existing Shannon entropy-based models are tested against twenty-one experimental datasets from the literature. The results show that the model of Kundu and the Shannon entropy-based expression have superior prediction accuracy with respect to experimental data compared with other models. With the exception of these models, the Tsallis entropy-based expression has the highest correlation coefficient value and the lowest root mean square error value for experimental data among the other models. This study indicates that the Tsallis entropy could be a good addition to existing deterministic models for predicting the lateral distribution of the velocity-dip position of rectangular open channel flow. This work also shows the potential of entropy-based expressions, the Shannon entropy and the Tsallis entropy in particular, to predict the velocity-dip position at the centerline of both narrow and wide rectangular open channels.

Assessment of urban flood susceptibility using semi-supervised machine learning model.

In order to identify flood-prone areas with limited flood inventories, a semi-supervised machine learning model-the weakly labeled support vector machine (WELLSVM)-is used to assess urban flood susceptibility in this study. A spatial database is collected from metropolitan areas in Beijing, including flood inventories from 2004 to 2014 and nine metrological, geographical, and anthropogenic explanatory factors. Urban flood susceptibility is mapped and compared using logistic regression, artificial neural networks, and a support vector machine. Model performances are evaluated using four evaluation indices (accuracy, precision, recall, and F-score) as well as the receiver operating characteristic curve. The results show that WELLSVM can better utilize the spatial information (unlabeled data), and it outperforms all comparison models. The high-quality WELLSVM flood susceptibility map is thus applicable to efficient urban flood management.

Modelling the Hindered Settling Velocity of a Falling Particle in a Particle-Fluid Mixture by the Tsallis Entropy Theory.

The settling velocity of a sediment particle is an important parameter needed for modelling the vertical flux in rivers, estuaries, deltas and the marine environment. It has been observed that a particle settles more slowly in the presence of other particles in the fluid than in a clear fluid, and this phenomenon has been termed 'hindered settling'. The Richardson and Zaki equation has been a widely used expression for relating the hindered settling velocity of a particle with that in a clear fluid in terms of a concentration function and the power of the concentration function, and the power index is known as the exponent of reduction of the settling velocity. This study attempts to formulate the model for the exponent of reduction of the settling velocity by using the probability method based on the Tsallis entropy theory. The derived expression is a function of the volumetric concentration of the suspended particle, the relative mass density of the particle and the particle's Reynolds number. This model is tested against experimental data collected from the literature and against five existing deterministic models, and this model shows good agreement with the experimental data and gives better prediction accuracy than the other deterministic models. The derived Tsallis entropy-based model is also compared with the existing Shannon entropy-based model for experimental data, and the Tsallis entropy-based model is comparable to the Shannon entropy-based model for predicting the hindered settling velocity of a falling particle in a particle-fluid mixture. This study shows the potential of using the Tsallis entropy together with the principle of maximum entropy to predict the hindered settling velocity of a falling particle in a particle-fluid mixture.

An Expression for Velocity Lag in Sediment-Laden Open-Channel Flows Based on Tsallis Entropy Together with the Principle of Maximum Entropy.

In the context of river dynamics, some experimental results have shown that particle velocity is different from fluid velocity along the stream-wise direction for uniform sediment-laden open-channel flows; this velocity difference has been termed velocity lag in the literature. In this study, an analytical expression for estimating the velocity lag in open-channel flows was derived based on the Tsallis entropy theory together with the principle of maximum entropy. The derived expression represents the velocity lag as a function of a non-dimensional entropy parameter depending on the average and maximum values of velocity lag from experimental measurements. The derived expression was tested against twenty-two experimental datasets collected from the literature with three deterministic models and the developed Shannon entropy-based model. The Tsallis entropy-based model agreed better with the experimental datasets than the deterministic models for eighteen out of the twenty-two total real cases, and the prediction accuracy for the eighteen experimental datasets was comparable to that of the developed Shannon entropy-based model (the Tsallis entropy-based expression agreed slightly better than the Shannon entropy-based model for twelve out of eighteen test cases, whereas for the other six test cases, the Shannon entropy-based model had a slightly higher prediction accuracy). Finally, the effects of the friction velocity of the flow, the particle diameter, and the particles' specific gravity on the velocity lag were analyzed based on the Tsallis entropy-based model. This study shows the potential of the Tsallis entropy theory together with the principle of maximum entropy to predict the stream-wise velocity lag between a particle and the surrounding fluid in sediment-laden open-channel flows.

Using Shannon entropy to model turbulence-induced flocculation of cohesive sediment in water.

Turbulence-induced flocculation of cohesive fine-grained sediment plays an important role in the transport characteristics of pollutants and nutrients absorbed on the surface of sediment in estuarine and coastal waters via the complex processes of sediment transport, deposition, resuspension and consolidation. In this study, the concept of Shannon entropy based on probability is applied to modelling turbulence-induced flocculation of cohesive sediment in water. Using the hypothesis regarding the cumulative distribution function, the function of floc size with flocculation time is derived by assuming a characteristic floc size as a random variable and maximizing the Shannon entropy, subject to certain constraints. The Shannon entropy-based model is capable of modelling the variation in floc size as the flocculation time progresses from zero to infinity. The model is tested against some existing experimental data from the literature and against a few deterministic mathematical models. The model yields good agreement with the observed data and yields better prediction accuracy than the other models. The parameter that has been incorporated into the model exhibits an empirical power-law relationship with the flow shear rate. An empirical model formulation is proposed, and it exhibits high prediction accuracy when applied to existing experimental data.

Changes in the two-dimensional and perimeter-based fractal dimensions of kaolinite flocs during flocculation: a simple experimental study.

In this study, Couette flow experiments were performed to estimate the temporal evolution of the 2D and perimeter-based fractal dimension values of kaolinite flocs during flocculation. The fractal dimensions were calculated based on the projected surface area, perimeter length and length of the longest axis of the flocs as determined by sampling observation and an image-processing system. The 2D fractal dimension, which relates the longest axis length and projected surface area of flocs, was found to decrease with the flocculation time, corresponding to the production of some porous flocs from the flow shear. This fractal dimension finally reached a steady state, which resulted from a dynamic equilibrium among the floc growth, floc breakage and floc restructuring. The perimeter-based fractal dimension, which characterizes the relationship between the projected surface area and the perimeter of flocs, increases with flocculation time because the flow shear increases the collisions among the primary particles, and some irregular flocs are formed. The perimeter-based fractal dimension reaches a steady level because of the balance among floc aggregation, breakage and restructuring. In addition, a stronger turbulent flow shear makes the steady state of fractal dimensions occur early during flocculation.