Journal of Hydrologic Engineering

A daily rainfall-runoff model has been improved by the integration of artificial neural network (ANN) and genetic algorithm (GA). The integrations are carried out on the daily rainfall-runoff model Génie rural à 4 paramètres journalier (GR4J). GR4J consists of production and routing storages. The production storage has only one process parameter and the routing storage has three. The ANN integration eliminates the three routing parameters. Automatic calibration capability has been added to the new hybrid model by integrating GA. The new hybrid model, which uses antecedent rainfall and temperature series, is applied to the Gediz River Basin in western Turkey. The results reveal that the hybrid model has better prediction performance than the original GR4J as well as the single ANN–based runoff prediction model.

The hydrometeorological method to estimate probable maximum precipitation (PMP) has several problems, one of which is related to the update of PMP. This study proposes using the Bayesian method to overcome this problem of updating the PMP. Because this is an amendment to the hydrometeorological method of estimating PMP, other procedures like moisture maximization and storm transposition remain unchanged. This study considers the Seoul rain gauge station in Korea as an application example. A total of 110 major storms that occurred in Korea between 1969 and 2017 are transposed to Seoul to be used in this study. The results are that first, reasonable PMPs are derived only when the likelihood function is based on a step increase function from zero to one at the transposed precipitation. Second, PMPs are updated only when a new, higher transposed precipitation than the previous transposed storms is considered. Third, the PMPs estimated in this study are found to be very similar to the PMPs reported by the Korean government in 1987, 2000, and 2004. However, the behavior of the PMPs in this study, i.e., the duration-PMP relation, is smoother than the PMPs reported by the Korean government. Based on these findings, it is concluded that the Bayesian method can be easily and effectively used for the update of PMPs. Also, an important advantage of using the Bayesian method is that it may substitute for the envelopment procedure in PMP estimation.

In the absence of measured streamflow, statistically simulated daily streamflow can be used to support the ability of physical models to represent hydrologic processes at ungauged locations. This study determined the feasibility of using statistical simulations in place of measured streamflow to calibrate physical models in ungauged basins. Daily streamflow was simulated at each of the 1,410 gauged watersheds using a cross-validated implementation of pooled ordinary kriging (POK). In this manner, the streamflow at each gauge was simulated as if no at-site streamflow information were available. The National Hydrologic Model application of the Precipitation-Runoff Modeling System was then calibrated through two separate procedures: (1) with measured streamflow, and (2) with statistically simulated streamflow in lieu of measured streamflow. Calibrating with statistically simulated streamflow produced performance within 23% of the performance of applications with knowledge of at-site measurements. Furthermore, statistically generated streamflow produced accurate timing information, which, when combined with alternative data sets (e.g., evapotranspiration, recharge, and so forth), can be used to improve representation of hydrologic processes at ungauged locations.

The Yangtze River is the third largest river basin in the world. With the advancement of research methods and data quality, the understanding of extreme climate changes in the Yangtze River Basin is constantly updated. This study used bias corrected climate projections fitted by stationary and nonstationary extreme generalized extreme value models to quantify historical warm, cold, and rainfall extremes in the Yangtze River Basin and their possible future changes. The uncertainty resulting from the intermodel and emission scenario differences was also discussed. The future annual maximum (minimum) temperature will possibly increase by 1.45°C–4.02°C (1.07°C–2.03°C), 1.54°C–4.45°C (0.99°C–1.76°C), and 1.60°C–4.91°C (0.84°C–1.54°C) in 2100 at the 10-, 20-, and 50-year return periods, respectively. The precipitation extremes are expected to increase by 6.4%–11.6%, 6.6%–12.5%, and 7.0%–14.6% at the 10-, 20-, and 50-year return periods, respectively. The warming trend and spatiotemporal distribution are mainly affected by monsoon climate, altitude, and ocean characteristics, and it is more pronounced in the middle and lower reaches of the river for warm extremes, but in plateau and coastal regions for cold extremes. The stationary generalized extreme value (GEV) model may lead to an overestimation of simulated temperature and precipitation extremes, with a deviation of $<1.0°\mathrm{C}$ for temperature extremes and $<10\%$ for precipitation extremes.

Decision making and hydrologic design for coping with floods are complex tasks in poorly gauged high-relief basins. The response of such basins is driven by precipitation and temperature, which controls the freezing level elevation and size of the runoff-contributing area. Moreover, early warning of floods based solely on real-time in situ monitoring is impractical. This study presents a meteorological characterization of daily flows based on off-site daily precipitation and temperature data in a high-relief catchment in central Chile. The results show that the variables that best explain daily discharges are the cumulative precipitation over the previous 3 days measured at a high elevation and the minimum temperature on the day of the maximum discharge measured at a lower elevation in the valley. These variables were used to build three multivariate regression models, based on principal component analysis, which are able to predict the occurrence of daily flows, particularly for low exceedance probabilities. Although developed for a particular catchment, and despite the specific meteorological threshold magnitudes identified for the catchment, the analysis is easily extendable to other similar high-relief locations.

This study investigates the contribution of tropical cyclones to the statistics of extreme rainfall events in Florida for durations ranging between 15 min and 7 days. It is shown that, at most rain-gauge locations, the probability distribution of annual maximum rainfall amounts associated with tropical cyclones is significantly different from the probability distribution of annual maximum rainfall amounts not associated with tropical cyclones. Typically, the probability distribution of the annual maximum rainfall can be better described by a mixed probability distribution rather than by the conventional lumped distribution, leading to differences in the estimated rainfall amounts associated with given return periods. The results presented in this paper indicate that more accurate intensity-duration-frequency functions could be obtained by separately accounting for rainfall associated with tropical cyclones.

Climate variability and human activities have substantially affected the delivery process of suspended sediment load (SSL) in rivers, which is a major driving force of river configuration and process in a basin. Thus, it is important to quantitatively identify how the SSL responds to environment changes. In this study, the Ganjiang River Basin (GRB), the main sediment source of Poyang Lake, China, is chosen to investigate how SSL changes under both climate variability and human activities. First, two nonparametric statistical methods, Mann-Kendall and Pettitt tests, are employed to detect the long-term trend and change point, respectively, of the SSL from 1958 to 2014. Then, the subbasins of GRB are divided according to the terrain slope gradient considering that the erosion rate of suspended sediments is probably influenced by terrain gradient. Effective annual areal rainfall above 12 different intensity threshold levels for different subbasins over the identified prechange period are investigated to identify the most effective rainfall vector affecting SSL in the multiple linear regression analysis. Finally, taking the prechange years as the baseline period, the impacts of environment changes on SSL during the postchange period are quantified using two methods (i.e., multiple linear regression and climate elasticity method). Benchmarked against the baseline period of 1958–1989, the mean annual SSL of GRB has decreased by 48% in 1990–2002, followed by 75% in 2003–2014. The consistent results of the two methods indicate that the impacts of human activities are the governing factor for the reduction of SSL, whereas climate variability enlarges the SSL due to the increase of areal rainfall. Furthermore, among the human-induced reductions of the SSL, about 47% (1990–2002) and 24% (2003–2014) are attributed to the sedimentation caused by the Wan’an Reservoir, the largest reservoir in the GRB, until 2014.