• Journal of The Korean Society of Agricultural Engineers
• /
• v.54 no.4
• /
• pp.127-135
• /
• 2012

This paper introduced the flow forecast modeling system that a water management agency in west central Florida, Tampa Bay Water has been operated to forecast monthly rainfall and streamflow in the Tampa Bay region, Florida. We evaluated current 1-year monthly rainfall forecasts and flow forecasts and actual observations to investigate the benefits of incorporating rainfall forecasts into monthly flow forecast. Results for rainfall forecasts showed that the observed annual cycle of monthly rainfall was accurately reproduced by the $50^{th}$ percentile of forecasts. While observed monthly rainfall was within the $25^{th}$ and $75^{th}$ percentile of forecasts for most months, several outliers were found during the dry months especially in the dry year of 2007. The flow forecast results for the three streamflow stations (HRD, MB, and BS) indicated that while the 90 % confidence interval mostly covers the observed monthly streamflow, the $50^{th}$ percentile forecast generally overestimated observed streamflow. Especially for HRD station, observed streamflow was reproduced within $5^{th}$ and $25^{th}$ percentile of forecasts while monthly rainfall observations closely followed the $50^{th}$ percentile of rainfall forecasts. This was due to the historical variability at the station was significantly high and it resulted in a wide range of forecasts. Additionally, it was found that the forecasts for each station tend to converge after several months as the influence of the initial condition diminished. The forecast period to converge to simulation bounds was estimated by comparing the forecast results for 2006 and 2007. We found that initial conditions have influence on forecasts during the first 4-6 months, indicating that FMS forecasts should be updated at least every 4-6 months. That is, knowledge of initial condition (i.e., monthly flow observation in the last-recent month) provided no foreknowledge of the flows after 4-6 months of simulation. Based on the experimental flow forecasts using the observed rainfall data, we found that the 90 % confidence interval band for flow predictions was significantly reduced for all stations. This result evidently shows that accurate short-term rainfall forecasts could reduce the range of streamflow forecasts and improve forecast skill compared to employing the stochastic rainfall forecasts. We expect that the framework employed in this study using available observations could be used to investigate the applicability of existing hydrological and water management modeling system for use of stateof-the-art climate forecasts.

• Journal of Korea Water Resources Association
• /
• v.50 no.11
• /
• pp.735-743
• /
• 2017

With growing concerns about ever-increasing anthropogenic greenhouse gas emissions, it is crucial to enhance preparedness for unprecedented extreme weathers that can bring catastrophic consequences. In this study, we proposed a stochastic framework that considers uncertainty in weather forecasts for flood analyses. First, we calibrated a simple rainfall-runoff model against observed hourly hydrographs. Then, using probability density functions of rainfall depths conditioned by 6-hourly weather forecasts, we generated many stochastic rainfall depths for upcoming 48 hours. We disaggregated the stochastic 6-hour rainfalls into an hourly scale, and input them into the runoff model to quantify a probabilistic range of runoff during upcoming 48 hours. Under this framework, we assessed two rainfall events occurred in Bocheong River Basin, South Korea in 2017. It is indicated actual flood events could be greater than expectations from weather forecasts in some cases; however, the probabilistic runoff range could be intuitive information for managing flood risks before events. This study suggests combining deterministic and stochastic methods for forecast-based flood analyses to consider uncertainty in weather forecasts.

• Water Engineering Research
• /
• v.5 no.2
• /
• pp.81-99
• /
• 2004

Long term streamflow regime under virtual climate change scenario was examined. Rainfall forecast simulation of the Canadian Global Coupled Model (CGCM2) of the Canadian Climate Center for modeling and analysis for the IPCC SRES B2 scenario was used for analysis. The B2 scenario envisions slower population growth (10.4 billion by 2010) with a more rapidly evolving economy and more emphasis on environmental protection. The relatively large scale of GCM hinders the accurate computation of the important streamflow characteristics such as the peak flow rate and lag time, etc. The GCM rainfall with more than 100km scale was downscaled to 2km-scale using the space-time stochastic random cascade model. The HEC-HMS was used for distributed hydrologic model which can take the grid rainfall as input data. The result illustrates that the annual variation of the total runoff and the peak flow can be much greater than rainfall variation, which means actual impact of rainfall variation for the available water resources can be much greater than the extent of the rainfall variation.

• Journal of Korea Water Resources Association
• /
• v.51 no.9
• /
• pp.813-826
• /
• 2018

Recently, weather radar system has been widely used for effectively monitoring near real-time weather conditions. The radar rainfall estimates are generally relies on the Z-R equation that is an indirect approximation of the empirical relationship. In this regards, the bias in the radar rainfall estimates can be affected by spatial-temporal variations in the radar profile. This study evaluates the uncertainty of the Z-R relationship while considering the rainfall types in the process of estimating the parameters of the Z-R equation in the context of stochastic approach. The radar rainfall estimates based on the Bayesian inference technique appears to be effective in terms of reduction in bias for a given season. The derived Z-R equation using Bayesian model enables us to better represent the hydrological process in the rainfall-runoff model and provide a more reliable forecast.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.12 no.1
• /
• pp.115-122
• /
• 1992

Stochastic rainfall-runoff process model which is mainly used in real-time streamflow forecasting is Transfer Function(TF) model that has a simple structure and can be easy to formulate state-space model. However, in order to forecast the streamflow accurately in real-time using the TF model, it is not only necessary to determine accurate structure of the model but also required to reduce forecasting error in early stage. In this study, after introducing 5-day Antecedent Precipitation Index (API5), which represents the initial soil moisture condition of the watershed, by using the threshold concept, the TF models in each API5 are identified by Box-Jenkins method and the results are compared with each other.