Journal of Hydrologic Engineering

Data of 13 stream flow gauging sites of the North Brahmaputra region of India are screened using the discordancy measure $\left(D_i\right)$ and homogeneity of the region is then tested employing the L-moments based heterogeneity measure $\left(H\right)$ . For computing heterogeneity measure $H$ , 500 simulations are performed using the Kappa distribution. Based on this test, it is observed that the data of 10 out of 13 gauging sites constitute a homogeneous region. Comparative regional flood frequency analysis studies are conducted employing the L-moments based commonly used frequency distributions. Based on the L-moment ratio diagram and $\mid Z_i^{\mathrm{dist}}\mid$ -statistic criteria, general extreme value (GEV) distribution is identified as the robust distribution for the study area. Regional flood frequency relationships are developed for estimation of floods of various return periods for gauged and ungauged catchments using the L-moment based GEV distribution and a regional relationship between mean annual peak flood and catchment area. Flood frequency estimates of gauged and ungauged catchments are compared; when, without meeting the criteria of regional homogeneity, data of all 13 gauging sites are used instead of data of only 10 gauging sites constituting the homogeneous region.

The southern oscillation index (SOI) series which is associated with El Nino was modeled as a linear stochastic model in the previous study. We also assume that it has a linear characteristic and is fitted to an autoregressive/moving average (ARMA) type model. The Bayesian information criterion is used for determining appropriate order of ARMA model and ARMA(1,8; 1) is chosen for the SOI series. The model is verified from the autocorrelation function, the partial autocorrelation function, and Porte Manteau test on the residuals for its validity. However, the hypothesis of randomness on the residual is rejected from a new test technique, called the Brock–Dechert–Scheinkman (BDS) statistic, which can detect the nonlinearity of time series that could not be determined by the conventional test techniques. This means that the ARMA model is not appropriate for the SOI series and this may be due to the nonlinearity of the time series. Therefore, we assume that the SOI series may have nonlinear properties and consider nonlinear modeling for the series. We use the close returns plot for searching for chaos, which has the nonlinear deterministic characteristics of a time series and found that there is no evidence of deterministic chaos in the SOI series. Therefore, we can consider that the nonlinear stochastic models may be more valid for the SOI series. The SOI series is fitted to the autoregressive conditional heteroscedasticity type model which has a nonlinear stochasticity and the model is tested on the residuals for its validity by the BDS statistic. As a result, the fitted nonlinear stochastic model is appropriate for the modeling of the SOI series and we may conclude that the nonlinear stochastic model is more valid for the SOI time series analysis and modeling than linear stochastic analog.

Goodness-of-fit statistics that accompany model calibration may not be good indicators of prediction accuracy. The difference between model calibration and validation accuracy has not been studied under the assumption that they are similar. Two methods of assessing prediction accuracy are compared. Split-sample testing and jackknifing are evaluated as a function of the sample size, the number of predictor variables, and the calibration accuracy. The relationship of split-sample testing is very sensitive to the sample size and the level of calibration accuracy. For small samples, split-sample testing is quite poor because of the loss of accuracy when the sample size is halved. The prediction accuracy decreases as the number of predictors increases because of the loss of degrees of freedom. Applications to regression analyses for a 10-year peak discharge prediction are provided. The results of an application to measured peak discharge data agree with simulation results. Future peak discharge studies should report both the calibration and jackknifing goodness-of-fit statistics.

This paper presents the results of a modeling study that is part of a collaborative distributed precipitation–snowmelt–runoff modeling study initiated by B.C. Hydro (the electric utilities company in British Columbia). Meteorological data were generated for a 23 year period by a high-resolution boundary layer model using a grid size of $2\min \text{latitude}\times 4\min \text{longitude}(3.7\times 4.7\mathrm{km})$ . The distributed daily precipitation and temperature–time sequences generated by this model were used as input to the WATFLOOD/SPL hydrological model, resulting in computed streamflows that were compared to measured streamflow at 32 flow gauging sites and four reservoirs. In this way, through a number of calibration and validation runs, the whole sequence of generating the meteorological data and their subsequent use to drive the hydrological model was tested. The study shows the importance of using both models in combination. Although the boundary layer model used the observed meteorological data to the fullest extent, the hydrologic simulations revealed a significant north–south trend in the precipitation error field. This resulted in streamflow errors as high as 60%. A reanalysis of the fields using the modeled streamflow reduced the error substantially. With the exception of a few streamflow stations, the computed flows matched the observed streamflows and reservoir inflows very well.

The linkage of surface runoff in urban areas to water quantity and quality has been studied extensively in the literature. Many computer models based on different routing theories have been developed to simulate the rainfall-runoff process in urban areas. Those computer models are classified into two categories: hydrologic models and hydraulic models. Hydrologic models are based on a hypothesized relation between outflow and water storage in the watershed, which is often modeled as a conceptual reservoir. The nonlinear reservoir method is a typical hydrologic routing method applied in models such as the Stormwater Management Model (SWMM). Hydraulic models are based on approximations of the real physical rainfall-runoff process. Kinematic-wave theory, utilized by many models including the Dynamic Watershed Simulation Model, is a commonly used hydraulic routing method. Previously unpublished experimental data obtained from the Watershed Experimentation System at the University of Illinois at Urbana-Champaign, Ill. were used to test the accuracy of these two routing methods. The nonlinear reservoir routing method as applied in SWMM may provide acceptable results for storms with durations longer than the watershed time of concentration (42 experiments) (average model fit efficiency $E=0.88$ ), but for storms with durations less than or equal to the time of concentration (26 experiments), poor results were obtained (average $E=0.07$ ). More accurate results generally were obtained using kinematic-wave routing (average $E=0.928$ and 0.807 for storm durations exceeding and not exceeding the watershed time of concentration, respectively). These results fit with the theoretical basis of the kinematic-wave theory that considers the actual physical processes in surface flow generation, while the nonlinear reservoir method does not consider the impact from the time lag needed for the flow depth to grow so that runoff can commence.

The problem of selecting the appropriate design flood is a constant concern to dam engineering and, in general, in the hydrological practice. Overtopping represents more than 40% of dam failures in the world. The determination of the design flood is based in some cases on the $T$ -year quantile of flood peak, and in other cases considering also the $T$ -year quantile of flood volume. However, flood peak and flood volume have a positive (strong or weak) dependence. To model properly this aspect a bivariate probability distribution is considered using the concept of 2-Copulas, and a bivariate extreme value distribution with generalized extreme value marginals is proposed. The peak–volume pair can then be transformed into the correspondent flood hydrograph, representing the river basin response, through a simple linear model. The hydrological safety of dams is considered checking adequacy of dam spillway. The reservoir behavior is tested using a long synthetic series of flood hydrographs. An application to an existing dam is given.

The development of a simple model for reconstructing the discharge hydrograph at a river site where only the stage is monitored and the discharge is recorded at another upstream section, is outlined. The model is designed to also accommodate significant lateral inflow contributions, permitting, without using a flood routing procedure and without the need of a rating curve at a local site, to relate the local hydraulic conditions with those at a remote gauged section. The approach is tested through its application to different flood events observed in three equipped river reaches of the Upper Tiber basin (Central Italy) where the contributing drainage area is significant. The simulated discharge hydrographs were found in a good agreement with those observed at the downstream ends of the investigated reaches. Errors in peak discharge and time to peak show that the model is suitable for estimating the rating curve beyond the range of the available streamflow measurements once an accurate stage–area relationship is assessed. Finally, the proposed approach was found more reliable than that based on the Muskingum method incorporating the water level observed at the downstream end.

A coupled surface/subsurface hydrologic model was developed to examine the effects of climatic conditions on stage fluctuations within a semipermanent wetland located in the Prairie Pothole region of east-central North Dakota. Model calibration was accomplished using data collected from 1981 until 1996 to encompass extreme climatic conditions. Simulation results suggest that wetland inflows are dominated by runoff (36%), direct precipitation to the lake (45%) and groundwater inflows (19%), while outflow from the wetland is primarily evaporation (84%) and groundwater discharge to the underlying aquifer (16%). Modeled surface runoff produces short duration, high magnitude flows typically associated with spring thaw while modeled groundwater provides flows smaller in magnitude but longer in duration that may be increasingly important to wetland stage during extended periods of drought and flood. Despite an oversimplification of the complex groundwater component of the wetland system it was found that this modeling approach was able to predict system response over $16\text{years}$ under extreme climatic conditions.

This study is designed to compare $1\text{day}$ ahead streamflow forecasting performance of multiple-layer perceptron (MLP) networks implemented at a daily time step for 47 watersheds spread across France and Central United States. In order to keep the task to manageable proportions, a large sample of test watersheds asks for a reduction of the number of steps in the network implementation procedure. This is achieved by eliminating the long trial and error process of input selection. Results show that it is feasible to obtain good $1\text{day}$ ahead streamflow forecasting performance from simple MLPs and input vectors consisting solely of the last observed streamflow and a predetermined range of precipitation observations that is roughly equal to the time of concentration of the watersheds. Also, intuitive preprocessing such as differencing the streamflow noticeably improves the forecasting performance in almost all instances. On the other hand, consideration of the potential evapotranspiration as an additional input decreases the MLP’s performance in the majority of instances. Finally, it is noteworthy that there is a general trend between the watershed runoff coefficients and the ability of the MLPs to correctly map $1\text{day}$ ahead streamflows.