Journal of Hydrologic Engineering

An experiment on predicting multivariate water resource time series, specifically the prediction of hydropower reservoir inflow using temporal neural networks, is presented. This paper focuses on dynamic neural networks to address the temporal relationships of the hydrological series. Three types of temporal neural network architectures with different inherent representations of temporal information are investigated. An input delayed neural network (IDNN) and a recurrent neural network (RNN) with and without input time delays are proposed for multivariate reservoir inflow forecasting. The forecast results indicate that, overall, the RNN obtained the best performance. The results also suggest that the use of input time delays significantly improves the conventional multilayer perceptron (MLP) network but does not provide any improvement in the RNN model. However, the RNN with input time delays remains slightly more effective for multivariate reservoir inflow prediction than the IDNN model. Moreover, it is found that the conventional MLP network widely used in hydrological applications is less effective at multivariate reservoir inflow forecasting than the proposed models. Furthermore, the experiment shows that employing only time-delayed recurrences can be the more effective and less costly method for multivariate water resources time series prediction.

Antecedent precipitation too often remains a subjectively determined and arbitrarily implemented parameter in rainfall-runoff modeling. Antecedent moisture can account for a several-fold difference in runoff from the same storm in an otherwise identical watershed. The antecedent precipitation index and the Natural Resource Conservation Service's antecedent moisture condition triad are conventional expressions of antecedent precipitation. Unfortunately, neither index consistently characterizes the runoff consequence of watershed moisture preceding a rainfall event. A normalized antecedent precipitation index is proposed that modifies the conventional index in three aspects: inclusion of antecedent precipitation earlier in the day of the event, normalization to the station mean, and normalization to the antecedent series length. The proposed index, based on daily rainfall record, is theoretically simple and mathematically continuous. Initial results show the proposed index to outperform single curve number-based results, even when the curve number is fit to historic rainfall-runoff records. The proposed index enhances delineation of the antecedent moisture condition for estimates based on that parameter and provides a basis for runoff estimates where the initial watershed moisture condition is probabilistic.

A new miscible seawater transport model, based on a variant form of the traditional method of characteristics (MOC), is presented. The distinctive features of this model are moving packets with preassigned volumes instead of concentration, numerical tracking of moving packets using fourth-order Runge-Kutta method, and direct computation of total transport due to advection and hydrodynamic dispersion. In addition to quantification of seawater circulation, the model output illustrates (1) the path followed by seawater inside the aquifer; (2) the buildup of seawater storage; and (3) the position and advance/retreat of a disperse interface. The model has been assessed using two benchmark test problems and subsequently applied to the Biscayne aquifer in Florida. All model runs were characterized by low mass balance errors (<0.2%).

Development of a new two-dimensional flood plain flow model, Hydro2de, is described. A finite volume approach is presented that ensures a nonoscillatory solution for flow conditions present on flood plains. The treatment of variable bed topography, friction losses including vegetation, and the boundaries between wet and dry cells are explained in detail. Tests are made to check conservation of mass for flows in the presence of initially dry domains. Numerical errors for flows diagonally across the calculation grid are analysed and a correction to reduce these errors is proposed. Part II outlines the performance of the model for two floods on the Waihao River flood plain.

Information from two floods on the Waihao River rural flood plain in New Zealand was used to validate a two-dimensional flood plain flow model, Hydro2de. An aerial photogrammetric survey digitally described the terrain. Measurements during and after the floods, and a global positioning system survey of positions from photographic records and other information recalled years later by flood plain residents, provided flood level, depth, and extent data to test the model. Uncalibrated, with inputs from the river and levee overtopping or breaches, the model underestimated levels, depths, and the area covered by floodwaters. The estimates were sufficiently close to be useful for flood plains without previous flood measurements. Calibrated to reproduce the measured flood extents, the model gave levels and depths closer to reality that were good at the edges of the flood plain subareas, but underestimated depths in the center of the flow. The underestimates occurred because the model did not include details of houses, hedges, and fences, nor any wave action of the water flow. A comparison is made to one-dimensional modeling using the MIKE11 model. Two-dimensional flood plain flow modeling promises benefits over one-dimensional modeling, in particular because the former does not require operator choice of a network of “channels” to represent the flood plain. The two-dimensional modeling reported here shows that improving the accuracy of the digital terrain model would provide the most improvement to accuracy of the results. Model functionality could also usefully be improved.

For a drainage design to be based on a consistent theory, the kinematic wave time of travel formula for channel flow ought to be used in conjunction with the kinematic wave overland time of concentration formula. This paper contains the kinematic wave time of travel formulas for channels of seven different cross sections. The channels are subject to a uniform lateral inflow and a constant upstream inflow. By means of a computer model that has been validated with the rainfall runoff data from a hydrologically isolated section of the M6 motorway in the United Kingdom, the times of travel in a channel with a vertical curb section were estimated for rainfall intensities ranging from 20 to 100 mm h⁻¹. A comparison of the time of travel estimates from the computer model, and from the time of travel formula, shows that the differences are less than 5%. The derived time of travel formulas are suitable for use in the design of drainage channels whose properties are within the applicability of the kinematic wave approximation.

An abrupt drawdown was observed at observation holes located in a geologic formation consisting of an interbedded sandstone-shale bedrock layers surrounding an underground liquefied petroleum gas cavern during operation. The transient hydraulic behavior of a multilayer system, such as interbedded sandstone-shale, is highly dependent on the hydraulic parameters of each individual layer. Therefore, in order to identify the cause of the drawdown, the changes in the hydraulic conductivity of each layer was studied based on compaction theory. It was observed that the compaction in shale occurs due to long-term lowered hydraulic head, and this causes the increase in equivalent hydraulic conductivity of the multilayers. The result of the numerical modeling showed that the altered hydraulic conductivity subsequently induced the additional drawdown.

The power distribution is a special case of the Weibull distribution. It can be derived as the distribution function of the low flows of a stream, when the recession curve is assumed to be exponential, using a theoretical result for the probabilities of maximum dry period lengths. The power distribution is found to have a better fit to the minimum flows of some streams than other two-parameter distributions, such as the Weibull and lognormal. In many streams, the distribution of the smaller low flows is different from that of the remaining data. This part can be fitted by a power distribution function whose parameters are estimated by LL-moments, which are L-moments computed with greater weights for smaller observations. Plotting positions and LL-moments are derived for the power distribution and are applied to data from nine sites. Analyses of data from a number of streams in Turkey and the United Kingdom have shown that the probability plot correlation coefficient test accepts the goodness of fit at very high significance levels when parameters are estimated using LL-moments.

This study evaluates how the change of land cover has an impact on soil-moisture-field evolution. Generally, soil moisture is known to be affected by regional climate, soil properties, and land cover, etc., among which the land cover may be the most crucial factor determining short-term and eventually long-term evolution of the soil moisture fields. This change is mainly due to deforestation and urbanization, which also affect the local soil climate. In this study, the impact of land cover change on soil-moisture-field evolution is investigated by use of a simple model of soil moisture dynamics described by Entekhabi and Rodriguez-Iturbe (1994). Statistical characterization of the soil moisture field, model parameter estimation, and example application are given utilizing the Washita '92 soil moisture data. Also, the Monsoon '90 data are used for implementing the temporal statistics of the soil moisture field. The impact of land cover change is evaluated through a simulation study by applying various parameter sets with different second-order statistics. As a result of the study, we found that (1) as the variability of land cover increases, the soil moisture field dries up faster, (2) the variability of the soil moisture field becomes highest during rainfall, but rapidly recovers its original variability as time elapses, and (3) the diffusion effect is limited to the period of surface runoff, and is also very small compared with that of the loss rate.

This technical note summarizes the results from daily rainfall data analysis with relation to annual rainfall amounts. By investigating the behavior of correlation coefficients between the annual rainfall amounts and the within-the-year cumulative frequency distribution functions of wet days with various threshold rainfall depths, it was found that (1) the annual rainfall amount is weakly correlated with the number of either dry or wet days—thus, the zero rainfall depth is not a proper threshold for the explanation of dry or wet years; and (2) there exist certain thresholds of daily rainfall depth (the climatological thresholds) with which the cumulative frequency of wet days can be well related with the annual rainfall amount. Also with this threshold, the wet and dry years can be seen divided clearly into separate groups of the cumulative frequency distribution functions, without any overlaps, which were common when using the zero rainfall depth as a threshold. Because it is found that there exist certain thresholds to be used for explaining the annual rainfall amount changes, these might also be used effectively for the analysis of possible changes in flood or drought patterns caused by global warming.