Journal of Hydrologic Engineering

Reservoir inflow forecast plays a crucial part in programming, development, operation, and management of water resource systems. To better reveal the complex properties of daily reservoir inflow, a clustered deep fusion (CDF) approach is proposed in this paper. First, variational mode decomposition (VMD) is used to decompose the daily reservoir inflow series into multiple modes, which are clustered into different sets by fuzzy $c$-means according to the Xie-Beni index in view of attribute domain. In each cluster, a deep autoencoder model (DAE) is developed for deep representations of the attributes in the deep domain. DAE outputs are finally fused at the synthesis domain into the forecasting results using random forest (RF). In this way, the inflow time series may be successively observed in the attribute domain, deep domain, and synthesis domain, which results in a clearer understanding of reservoir inflow trend. The present approach is modeled and evaluated using historical data collected from the Three Gorges Reservoir, China. For comparison, two kinds of learning patterns—deep learning (VMD-DAE-RF and DAE) and shallow learning (feed-forward neural network, least-squares support regression, and RF)—are applied to the same case. The results indicate that the proposed CDF model outperforms all comparison models in terms of mean absolute percentage error (6.174%), root mean-square error ($\mathrm{1,077.428}{\mathrm{m}}^3/\mathrm{s}$), and correlation coefficient criteria (0.987). Thus, it is concluded that deep learning in the cluster fusion architecture is more promising.

Recently, human intervention and climate change have been proposed as the causes of changes in extreme water levels, which impact the likelihood of flooding, especially in coastal areas. In many studies, frequency analysis of extreme water levels has been conducted under nonstationary conditions, in which the parameters of a given distribution vary with time or several climatological variables. However, water levels possess unique characteristics in that they are strongly impacted by local influences, so the covariates used for nonstationary extreme water level modeling should be chosen with respect to the area of interest. For these reasons, it is important to consider local variables that have strong physical associations with flooding in the study of nonstationary extreme water levels. This study uses four local covariates, i.e., rainfall, sea level, urbanization growth, and outflows from upstream reservoirs to develop 92 nonstationary extreme water level models. The stationary models are also developed for comparison purposes. The results indicate that the nonstationary approach using local covariates is suitable for modeling extreme water levels in Ho Chi Minh City. Additionally, based on the best chosen statistical models, a significant influence of sea level and urbanization on nonstationarity in extreme water levels was found at all surveyed stations. Moreover, it was found that the extreme water levels derived from the stationary models were underestimated relative to the best nonstationary models for all stations, with the differences in magnitude reaching approximately 47%, 35%, 31%, and 25% at the Bienhoa, Nhabe, Phuan, and TDM stations, respectively.

This paper proposes, systematizes, and validates a methodology for stochastically generating streamflow time series at virtually any time scale. Starting from deseasonalized data (i.e., observed series removed of their periodicity), the proposed methodology works by sequentially sampling each new streamflow value from a joint probability density function (PDF) conditioned by the previously generated values. This joint PDF is obtained directly from the observed data, and it represents the probability distribution and probabilistic dependency between consecutive streamflow values. An example application of the series generation methodology was developed based on daily streamflow data from Portugal. The proposed methodology’s results showed good agreement between the observed and generated PDFs. Generated series display less than 10% and 25% deviation, respectively, in terms of the means and standard deviation (for the 0th, 1st, and 2nd order) and the serial autocorrelation, from the observed series. This provides strong evidence for the methodology’s capability for reproducing the streamflow’s autocorrelation structure.

The Soil and Water Assessment Tool (SWAT) is a well-established modeling tool for simulating hydrologic processes in watershed systems. However, SWAT often does not perform well in highly managed watersheds with intensive irrigation due to management practices and associated hydrologic processes (e.g., canal seepage) not included in the modeling code. This study presents a method to apply SWAT to managed irrigated watersheds, which includes (1) designating each cultivated field as a hydrologic response unit (HRU) and including recorded crop rotations; (2) applying scheduled irrigation according to water rights; and (3) simulating seepage from earthen irrigation canals, with the latter requiring a minor modification to the SWAT code. SWAT’s autoirrigation function is also used as a comparison. The methodology was applied to a $732\text{-}{\mathrm{km}}^2$ watershed in the Lower Arkansas River Valley (southeastern Colorado), a semiarid region that has been intensively irrigated for over 100 years. The model was tested against monthly stream discharge at five stream gauging stations in the Arkansas River and its tributaries during the 2001–2007 period. Results demonstrate the necessity of including human management features, particularly canal seepage, to capture in-stream flows in the river and tributaries. The scheduled irrigation approach has a significantly better performance in the smaller highly agricultural tributaries whereas the autoirrigation functionality has a slightly but not significantly better performance at the outlet of the whole watershed. The methods presented herein can be applied to other semiarid irrigated regions.

The sustainability of worldwide reservoirs is threatened by the reduction of their storage capacity caused by continuous sediment accumulation. Many reservoirs are filling in at rates higher than projected, and the storage initially allocated for sediment retention is no longer suitable. Only a limited number of sediment management strategies, such as watershed conservation or dam rehabilitation, can be used to restore capacities of large storage reservoirs. They generally require long-term planning and implementation that rely on accurate sedimentation estimates. This study develops a stochastic sediment budget approach based on three main variables (sediment delivery, trapping efficiency, and sediment dry bulk density) to estimate sedimentation in Fort Cobb Reservoir, a sparsely measured $98\text{-}{\mathrm{Mm}}^3$ storage reservoir located in western Oklahoma. To account for missing sediment delivery data, a number of temporal and spatial data expansion techniques based on historical records and spatial proximity are developed and tested. The budget approach provides estimates of sedimentation volumes in Fort Cobb Reservoir similar to results of sedimentation surveys for the 1959–1993 and 1993–2007 periods. However, these estimates are highly uncertain. Corresponding coefficients of variation are 86% for the poorly monitored 1959–1993 period, and 38% for the 1993–2007 period with the best available data. For the 1993–2007 period, dry bulk density, sediment loads, and trap efficiency contributed to 55%, 35%, and 10% of the volume uncertainty, respectively. Using the stochastic budget approach, it is estimated that 88% of the reservoir volume allocated to sediment storage will be filled by the end of its design life.