Failure Prediction of Municipal Water Pipes Using Machine Learning Algorithms

Pipe failure prediction has become a crucial demand of operators in daily operation and asset management due to the increase in operation risks of water distribution networks. In this paper, two machine learning algorithms, namely, random forest (RF) and logistic regression (LR) algorithms are employed for pipe failure prediction. RF algorithm consists of a group of decision trees that predicts pipe failure independently and makes the final decision by voting together. For the LR algorithm, the mapping relationship between existing data and decision variables is expressed by the logistic function. Then, the prediction is made by comparing the conditional probability with the fixed threshold value. The proposed algorithms are illustrated using an actual water distribution network in China. Results indicate that the RF algorithm performs better than the LR algorithm in terms of accuracy, recall, and area under the receiver operating characteristic curve. The effects of seven characteristics on pipe failures are analyzed, and diameter and length are identified as the top two influential factors.

Graphical Abstract

     

Unraveling the Lagged Effect of Hydro-meteorological Conditions On the Trophic State of a Reservoir By Applying Dynamic Regression

In this study we develop a novel approach to quantify the relative importance of hydro-meteorological (HM) conditions on the trophic state index (TSI) of a water reservoir (San Roque, Córdoba, Argentina). Seven HM variables measured at four reservoir sites and different depths over a time period of near 2 decades are used. We propose a dynamic regression model to predict the TSI from these variables aggregated over a range of time lags, which has not been applied in such a complex setting so far. By performing coefficient analysis, we quantify the relative importance of these variables on the TSI, as well as the time duration over which they have significant impact (lagged effect). Additionally, the analysis of the autoregressive and moving average (ARIMA) terms reveals the impact of the residual effects of previous trophic states on the current trophic state. We find that surface temperature and precipitation have the largest direct relationship to the TSI in the short-term, while the reservoir water level is inversely related to the TSI in the short- to mid-term. Also, the residual effects of the trophic state impact from 1 month (generally) up to 2 years (exceptionally). This approach can be applied to other water bodies affected by similar eutrophication phenomena.

Graphical abstract

     

Sensor Partitioning Placements via Random Walk and Water Quality and Leakage Detection Models within Water Distribution Systems

A novel sensor partitioning placement model is presented to evenly distribute sensors to water distribution systems (WDS) for monitoring leakages and contamination. First, random walk community detection (RWCD) is used to divide WDS into different partitions. Then, an extended period leakage detection (EPLD) model is presented. The total leakage detection and the average time of leakage detection are used as objective functions for pressure sensor placement. Next, the extended period water quality detection (EPWQD) model is presented. The total intrusion detection, the average percentage of clean water, and the average time of water quality detection are used as objective functions for water quality sensor placement. Evolutionary algorithm (EA) modules are applied to optimize the locations of pressure and water quality sensors. Seven networks are employed to verify the practicability of the model. The results show that leakage and intrusion detection rate is up to 85% during 24 h, and the average percentage of clean water is up to 0.9 in these cases. Finally, the model compares the leakage zone identification (LZI) and the water quality sensor placement strategy (WQSPS) models. The total detection number, the total average time of detection, and the total average percentage of clean water have been improved. Therefore, this model is a high-potential way of sensor placement.

Graphical Abstract

     

Random Walks Partitioning and Network Reliability Assessing in Water Distribution System

A novel network partition model is presented within the water distribution system (WDS). Firstly, random walk community detection (RWCD) is employed to divide WDS into different partitions concerning the average pressure of nodes. Then, network reliability is assessed based on hydraulic reliability estimation (HRE), mechanical reliability estimation (MRE), flow entropy function (FEF), and network resilience (NR), via optimizing boundary pipes by the non-dominated sorting genetic algorithm-II (NSGA-II). Finally, pressure-reducing valves (PRVs) are set to pipes for acquiring optimized partitions. The Open Water Analytics (OWA) toolbox and Matlab-2018b is adopted as a hydraulic calculation tool for these extended period simulations (EPS). Seven cases of WDSs were used to verify the practicability of this model. The results demonstrate that network reliability is improved effectively after partitioning and optimizing.

Graphical Abstract

     

Climate Variability Impact on the Spatiotemporal Characteristics of Drought and Aridityin Arid and Semi-Arid Regions

Investigating the spatiotemporal distribution of climate data and their impact on the allocation of the regional aridity and meteorological drought, particularly in semi-arid and arid climate, it is critical to evaluate the climate variability effect and propose sufficient adaptation strategies. The coefficient of variation, precipitation concentration index and anomaly index were used to evaluate the climate variability, while the Mann-Kendall and Sen’s slope were applied for trend analysis, together with homogeneity tests. The aridity was evaluated using the alpha form of the reconnaissance drought index (Mohammed & Scholz, Water Resour Manag 31(1):531–538, 2017c), whereas drought episodes were predicted by applying three of the commonly used meteorological drought indices, which are the standardised reconnaissance drought index, standardized precipitation index and standardized precipitation evapotranspiration index. The Upper Zab River Basin (UZRB), which is located in the northern part of Iraq and covers a high range of climate variability, has been considered as an illustrative basin for arid and semi-arid climatic conditions. There were general increasing trends in average temperature and potential evapotranspiration and decreasing trends in precipitation from the upstream to the downstream of the UZRB. The long-term analysis of climate data indicates that the number of dry years has temporally risen and the basin has experienced succeeding years of drought, particularly after 1994/1995. There was a potential link between drought, aridity and climate variability. Pettitt’s, SNHT, Buishand’s and von Neumann’s homogeneity test results demonstrated that there is an evident alteration in the mean of the drought and aridity between the pre- and post-alteration point (1994).

     

Digital Corporate Social Responsibility Reporting in the Water Industry

Companies in the Water Industry present digital Corporate Social Responsibility (CSR) agenda and, also, the social and environmental commitment to their stakeholders through the websites. The purpose of this research is to assess the digital CSR in Portuguese companies of the Water Industry. Furthermore, the research examines factors that impacts on the digital status of the online disclosure. The authors analyze the CSR information published on their websites of the Portuguese companies, operating in bottle water industry using empirical analysis. The data was collected based on the Global Reporting Initiative (GRI 2021a) standards that details the level of disclosure in this industry and highlight areas of underreporting. The results point to factors that need to improve to companies’ digital CSR report good practices and weak points based on the companies’ size, number of employees and turnover as factors that influence this level of disclosure.

     

Optimal flood control in multireservoir cascade systems with deterministic inflow forecasts

This article presents the formal analysis of a problem of the optimal flood control in systems of serially connected multiple water reservoirs. It is assumed, that the basic goal is minimization of the peak flow measured at a point (cross-section) located downstream from all reservoirs and that inflows to the system are deterministic. A theorem expressing sufficient conditions of optimality for combinations of releases from the reservoirs is presented together with the relevant proof. The main features of the optimal combinations of controls are thoroughly explained. Afterwards, two methods of determining the optimal releases are presented. Finally, the results of the application of the proposed methodology to a small, four reservoir system are presented.Notations c _i contribution of theith,i=1, ...,m, reservoir to the total storage capacity of the multireservoir system - d _i (t) one of the uncontrolled inflows to the cascade at timet (fori=1 main inflow to the cascade, fori=2, ...,m, side inflow to theith reservoir, fori=m+1 side inflow at pointP) - total inflow to theith reservoir,i=2, ...,m, at timet (i.e., inflowd _i augmented with properly delayed releaser _i–1 from the previous reservoir) (used only in figures) - d(t),d _S (t) (the first term is used in text, the second one in figures) aggregated inflow to the cascade (natural flow at pointP) at timet - time derivative of the aggregated inflow at timet - i reservoir index - m number of reservoirs in cascade - P control point, flood damage center - minimal peak of the flow at pointP (cutting level) - Q _p (t) flow measured at pointP at timet - flow measured at pointP at timet, corresponding to the optimal control of the cascade - r _i (t) release from theith reservoir at timet, i=1, ...,m - optimal release from theith reservoir at timet, i=1, ...,m - r ₁ ^* (t) a certain release from theith reservoir at timet, different than ,i=1, ...,m, (used only in the proof of Theorem 1) - a piece of the optimal release from themth reservoir outside period at timet - assumed storage of theith reservoir at time (used only in the proof of Theorem 1) - s _i (t) storage of theith reservoir at timet, i=1, ...,m - time derivative of the storage of theith reservoir at timet, i=1, ...,m - storage capacity of theith reservoir,i=1, ...,m - (the first term is used in text, the second one in figures) total storage capacity of the cascade of reservoirs - S* sum of storages, caused by implementingr _i ^* ,i=1, ...,m, of all reservoirs measured at (used only in the proof of Theorem 1) - t time variable (continuous) - t ₀ initial time of the control horizon - t _a initial time of the period of constant flow equal at pointP - initial time of the period of the essential filling of theith reservoir,i=1, ...,m (used only in the proof of Theorem 1) - t _b final time of the period of constant flow equal at pointP - final time of the period of the essential filling of theith reservoir,i=1, ...,m (used only in the proof of Theorem 1) - time of filling up of theith reservoir while applying method with switching of the active reservoir - t _f final time of the control horizon - fori=1, ...,m–1, time lag betweenith andi+1th reservoir; fori=m time lag between the lowest reservoir of the cascade and the control pointP     

Simulating Future Groundwater Recharge in Coastal and Inland Catchments

Groundwater is a primary source of drinking water in the Mediterranean, however, climate variability in conjunction with mismanagement renders it vulnerable to depletion. Spatiotemporal studies of groundwater recharge are the basis to develop strategies against this phenomenon. In this study, groundwater recharge was spatiotemporally quantified using the Soil and Water Assessment Tool (SWAT) in one coastal and one inland hydrological basin in Greece. A double calibration/validation (CV) procedure using streamflow data and MODIS ET was conducted for the inland basin of Mouriki, whereas only ET values were used in the coastal basin of Anthemountas. Calibration and simulation recharge were accurate in both sites according to statistical indicators and previous studies. In Mouriki basin, mean recharge and runoff were estimated as 16% and 9%, respectively. In Anthemountas basin recharge to the shallow aquifer and surface runoff were estimated as 12% and 16%, respectively. According to the predicted RCP 4.5 and 8.5 scenarios, significant variations in groundwater recharge are predicted in the coastal zone for the period 2020–2040 with average annual recharges decreasing by 30% (RCP 4.5) and 25% (RCP 8.5). Variations in groundwater recharge in the inland catchment of Mouriki were insignificant for the simulated period. Anthemountas basin was characterized by higher runoff rates. Groundwater management in coastal aquifers should include detailed monitoring of hydrological parameters, reinforced groundwater recharge during winter and reduced groundwater abstraction during summer depending on the spatiotemporal distribution of groundwater recharge.

     

Chemical Evaluation of Ma’an Sewage Effluents and its Reuse in Irrigation Purposes

The present study investigates the chemical composition of Ma’an Wastewater Treatment Plant in south Jordan. Samples of effluent of this plant were collected over 1 year period. All samples were analyzed for pH, conductivity, major ions (Cl⁻, , , , , Na⁺, K⁺, Ca^2 + and Mg^2 +) and trace metals B, Fe, Cu, Zn, Cd and Pb. The pH value ranges from 6.79 to 8.15 with a median value of 7.39 ± 0.32. The water quality was characterized by its high salinity hazard (C3) and low sodium hazard (S1) which can be considered as marginal for human consumption. Moreover, concentrations of trace metals in treated wastewater were found to be low and within guidelines for irrigation water due to low level of industrialization activities in the study area. Generally, the result of this study suggests that the treated wastewater is suitable for irrigational purposes, while these effluents can be considered as possible additional resources for irrigation in Jordan.     

Evaluation of Hantush’s S Function Estimation Methods for Predicting Rise in Water Table

Hantush’s model is widely used for predicting rise in water table in response to groundwater recharge. Several approximate methods of the Hantush mound function, S(α, β) have been developed to overcome the limitations of Hantush’s tabulated values of the S(α, β) function. These approximate methods have their own advantages and disadvantages, and it is difficult to identify the most accurate and computationally efficient S(α, β) estimation method. In this study, performance of four different algebraic approximate S(α, β) estimation methods are compared with the Hantush method using the published data. The four different methods considered are Swamee and Ohja (1997) (SO), Singh (2012) (SI), Vatankhah (2013) (VA) and Gauss-Legendre quadrature (GL) method with various Gaussian points (GP). Seven statistical accuracy and computation efficiency indicators are used to assess the performance of different S(α, β) estimation methods. The GL method with 100 to16 GPs is found to be the most accurate S(α, β) estimation method. This is followed by the SO, GL with 14 to 12 GPs, VA, GL with 10 GP, SI, and GL with 9 to 3 GPs. A good trade off between accuracy and efficiency is found with the SO, VA, and GL method with 14, 12 and 10 GPs. Comprehensive analysis of different S(α, β) estimation methods, and their ranking based on overall performance index will be helpful in modelling water table rise due to groundwater recharge, optimum design of recharge basin, and evaluation of effectiveness of recharge basins in groundwater recharging.

     

Calibration Procedure of Regional Flow Duration Curves Evaluating Water Resource Withdrawal from Diversion Dams

In the last decades, climatic changes in Mediterranean regions and frequent events of water resource scarcity in supply systems have required addressing the problem of increasing the inflows in storage reservoirs by connecting them to diversion dams. Usually, diversion dams do not have a large storage volume; consequently, these works are not able to regulate monthly flow, and they can divert to reservoirs only a part of river flow. In this field of research, this study aims to provide a procedure to evaluate water volumes withdrawn from diversion dams considering river flows and transfer flows. The procedure is founded by analysing 36 gauging stations in Sardinia (Italy), where a unique regional flow duration curve (FDC) can be defined. The effectiveness of regional evaluation of volume withdrawn, based on a monthly time scale analysis, has been investigated. The methodology allows finding optimal values of regional FDC parameters in order to better evaluate water resources withdrawn from diversion dams with respect to the current estimation used in the Sardinia Region Water Plan (SRWP). The current SRWP (RAS 2006) uses a fixed and extremely precautionary value of the FDC that underestimates withdrawal volumes. Moreover, a correlation analysis has been carried out in order to extend the evaluation of optimal FDC parameters for ungauged basins that allows to improve application of this procedure. Obtained results could allow to update the SRWP as highlighted in the final application to a real water system.

     

Coupling Singular Spectrum Analysis with Least Square Support Vector Machine to Improve Accuracy of SPI Drought Forecasting

The study proposed a Standardized Precipitation Index (SPI) drought forecasting model based on singular spectrum analysis (SSA) and single least square support vector machine (LSSVM) with a twofold investigation: (Beguería et al. Int J Climatol, 34(10): 3001–3023, 2014) the forecasting performance of the LSSVM-based model with or without coupling SSA and (Belayneh et al. J Hydrol, 508: 418–429, 2014) the model performances by using different inputs (i.e., antecedent SPIs and antecedent accumulated monthly rainfall) preprocessed by SSA. For the first part investigation, the LSSVM-based model using antecedent SPI as input (LSSVM1) and the LSSVM-based model coupling with SSA using antecedent SPI as input (SSA-LSSVM2) were developed. For the second part of investigation, the SSA-LSSVM-based model using antecedent accumulated monthly rainfall as input (SSA-LSSVM3) was developed and compared to SSA-LSSVM2. The drought indices (SPI3 and SPI6) were chosen as the outputs of the SPI drought forecasting models. The Tseng-Wen reservoir catchment in southern Taiwan was selected to test the aforementioned models. The results show that the forecasting performance of SSA-LSSVM2 is better than that of LSSVM1, which means the input data preprocessed by SSA can significantly increase the accuracy of the SPI drought forecasting. In addition, the performance comparison between SSA-LSSVM2 and SSA-LSSVM3 indicates that using antecedent accumulated monthly rainfalls (i.e., 3-month and 6-month accumulated rainfalls) as input of SSA-LSSVM3 are much better than using antecedent SPIs (i.e., SPI3 and SPI6) as input of SSA-LSSVM2. SSA-LSSVM3 is found to be the most appropriate model for SPI drought forecasting in the case study.

     

Improved Water Allocation under Limited Water Supplies Using Integrated Soil-Moisture Balance Calculations and Nonlinear Programming

Water allocation under limited water supplies is becoming more important as water becomes scarcer. Optimization models are frequently used to provide decision support to enhance water allocation under limited water supplies. Correct modelling of the underlying soil-moisture balance calculations at the field scale, which governs optimal allocation of water is a necessity for decision-making. Research shows that the mathematical programming formulation of soil-moisture balance calculations presented by Ghahraman and Sepaskhah (2004) may malfunction under limited water supplies. A new model formulation is presented in this research that explicitly models deep percolation and evapotranspiration as a function of soil-moisture content. The new formulation also allows for the explicit modelling of inefficiencies resulting from nonuniform irrigation. Modelling inefficiencies are key to the evaluation of the economic profitability of deficit irrigation. Ignoring increasing efficiencies resulting from deficit irrigation may render deficit irrigation unprofitable. The results show that ignoring increasing efficiencies may overestimate the impact of deficit irrigation on maize yields by a maximum of 2.2 tons per hectare.

     

Surface Water Quantity for Drinking Water during Low Flows - Sensitivity Assessment Solely from Climate Data

The future sensitivity of the surface water supply of Québec City is assessed in this paper using two methodologies: the methodology that has prevailed since the publication of the AR4 report, the hydroclimatological modeling framework, and an alternative approach adapted from Foulon et al. (2018). This alternative approach captures past relationships between climate data indices (CDIs), such as cumulative rainfall, and hydrological data indices (HDIs), such as 7-day low flows, and applies these relationships to assess future trends. Future climates were built for two emission scenarios, RCP-4.5 and???8.5, and the uncertainty of climate change was addressed through the use of 16 climate models. Overall, both methodological frameworks predicted similar low flow trends for the reference and future horizons (2016–2045 and 2046–2075). The future pressure on the surface water supply of Québec City should raise concerns. Indeed, for RCP-8.5, results indicated a decrease in the PI₁ values (ratio of 2-year low flow to water abstraction rate) of around 20% (2016–2045) and 35% (2046–2075) with a fairly high confidence (around 90% of models agreeing on the direction of change); leading to values less than 1; indicating an insufficient water supply with respect to available water during 2-year low flows. These results demonstrate the capacity of the method to provide a screening assessment of future drought-prone-watersheds. Furthermore, the application of the alternative approach, given climate simulations, would help early implementation of good management practices even for municipalities that do not have the capacities to conduct the more conventional approach.

     

Comparative Evaluation of Analytic Solutions in Predicting Soil Moisture Profiles in Vertical One-Dimensional Infiltration under Ponded and Constant Flux Boundary Conditions

The problem of vertical one-dimensional infiltration for both ponded and constant flux boundary conditions was studied through the use of existing analytic solutions. Main objective was to compare the soil moisture profile developed under constant flux boundary condition at the time of ponding , with that moisture profile developed under ponded conditions at an earlier time . Time t _C denotes the time when the decreasing infiltration rate for the ponded conditions becomes equal to the constant flux _q , applied for the constant flux case. One might state that the analytical solutions, for both cases do not give identical profiles. An approximate coincidence might be brought about through a modification in the diffusivity which, in many respects, seems justified. Practical outcome of the above analysis is the determination of the time of ponding T, after which surface runoff starts, for the constant flux case. This is of practical significance either under natural conditions of rainfall or under conditions of sprinkle-irrigation, since surface runoff is directly related with soil erosion and waste of irrigation water. Therefore any attempt to determine the time of ponding T is well merited.     

A real-time flood forecasting model based on maximum-entropy spectral analysis: II. Application

The MESA-based model, developed in the first paper, for real-time flood forecasting was verified on five watersheds from different regions of the world. The sampling time interval and forecast lead time varied from several minutes to one day. The model was found to be superior to a state-space model for all events where it was difficult to obtain prior information about model parameters. The mathematical form of the model was found to be similar to a bivariate autoregressive (AR) model, and under certain conditions, these two models became equivalent.Notation A _k parameter matrix of the bivariate AR model - B backshift operator in time series analysis - eT forecast error (vector) at timet = T - _t uncorrelated random series (white noise) - F _k forward extension matrix of the entropy model forkth lag - I identity matrix - m order of the entropy model - N number of observations - P order of the AR model - Q _p peak of the direct runoff hydrograph - R correlation matrix - t _p time to peak of the direct runoff hydrograph - ₁ coefficient of variation - ₂ ratio of absolute error to the mean - forecasted runoff - x _i observed runoff - mean of the observed runoff - X ^–1 inverse ofX matrix - X* transpose of theX matrix Abbreviations AIC Akaike information criterion - AR autoregressive (model) - AR(p) autoregressive process of thepth order - ARIMA autoregressive integrated moving average (model) - acf autocorrelation function - ccf cross-correlation function - FLT forecast lead time - MESA maximum entropy spectral analysis - MSE mean square error - STI sampling time interval     

A real-time flood forecasting model based on maximum-entropy spectral analysis: I. Development

This paper, the first of two, develops a real-time flood forecasting model using Burg's maximum-entropy spectral analysis (MESA). Fundamental to MESA is the extension of autocovariance and cross-covariance matrices describing the correlations within and between rainfall and runoff series. These matrices are used to derive the model forecasting equations (with and without feedback). The model may be potentially applicable to any pair of correlated hydrologic processes.Notation a _k extension coefficient of the model atkth step - B _k backward extension matrix forkth step - B _ijk element of the matrixB _k (i,j=1, 2) - c _k coefficient of the entropy model atkth step in the LB algorithm - e _k (e _x ,e _y )k = forecast error vector atkth step - E _k error matrix atkth step - E _ijk element of theE _k (i,j=1, 2) - f frequency - F _k forward extension matrix atkth step - F _ijk element of theF _k matrix (i,j=1, 2) - H(f) entropy expressed in terms of frequency - H _X entropy of the rainfall process (X) - H _Y entropy of the runoff process (Y) - H _XY entropy of the rainfall-runoff process - I identity matrix - forecast lead time - m model order, number of autocorrelations - R correlation matrix - S _x standard deviation of the rainfall data - S _y standard deviation of the runoff data - t time - T ₁ rainfall record - T ₂ runoff record - T rainfall-runoff record (T=T ₁ T ₂) - x _t rainfall data (depth) - X X() = rainfall process - mean of the rainfall data - y _t direct runoff data (discharge) - Y Y() = runoff process - mean of the runoff data - (x, y) _t rainfall-runoff data (att T) - (x, y, z) _t rainfall-runoff-sediment yield data (att T) - z complex number (in spectral analysis) - _k coefficient of the LB algorithm atkth step - _nj Lagrange multiplier atjth location in the _n matrix - _{n n} = matrix of the Lagrange multiplier atkth step - _X (k), _Y (k) autocorrelation function of rainfall and runoff processes atkth lag - _XY (k) cross-correlation function of rainfall and runoff processes atkth lag - W ₁(f) power spectrum of rainfall or runoff - W ₂(f) cross-spectrum of rainfall or runoff Abbreviations acf autocorrelation function - ARMA autoregressive moving average (model) - ARMAX ARMA with exogenous input - ccf cross-correlation function - det() determinant of the (...) matrix - E[...] expectation of [...] - FLT forecast lead time - KF Kalman filter - LB Levinson-Burg (algorithm) - MESA maximum entropy spectral analysis - MSE mean square error - SS state-space (model) - STI sampling time interval - forecast ofx - forecast ofx -step ahead - x _F feedback ofx-value (real value) - |x| module (absolute value) ofx - X ^–1 inverse of the matrixX - X* transpose of the matrixX     

An Improved I a S Relation Incorporating Antecedent Moisture in SCS-CN Methodology

Employing a large dataset of 84 small watersheds (area = 0.17 to 71.99 ha) of U.S.A., this paper investigates a number of initial abstraction (I _a )-potential maximum retention (S) relations incorporating antecedent moisture (M) as a function of antecedent precipitation (P ₅), and finally suggests an improved relation for use in the popular Soil Conservation Service Curve Number (SCS-CN) methodology for determination of direct runoff from given rainfall. The improved performance of the incorporated M = α and I _a = λ S ²/(S + M) relations, where λ is the initial abstraction coefficient, in the SCS-CN methodology exhibits the dependence of I _a on M, which is close to reality; the larger the M, the lesser will be I _a , and vice versa. Such incorporation obviates sudden jumps in the curve number variation with antecedent moisture condition, an unreasonable and undesirable feature of the existing SCS-CN model.