Hydrologic Time Series Analysis I Posters
Presiding: P Rasmussen, University of Manitoba; S Durrans, University of Alabama
H43A-01 1330h
Simulation of Monthly Flows Using a Markov Switching Model
Some annual time series of streamflow and precipitation exhibit extended wet and dry spells that may not be well captured by low-order ARMA models. Markov Switching (MS) models have been suggested as a possible alternative to traditional time series models. MS models employ a Markov chain to simulate a state variable representing the climate regime. Within each climate regime, the hydrologic variable is described by a probability density function whose parameters depend on the state. To simulate realistic multi-year flow regimes, the MS model must be applied to annual data. However, in typical water resource studies, synthetic streamflow data are needed at sub-annual time steps, for example monthly. We present a stochastic model that adapts the well known disaggregation approach to the case of MS models. Parameter estimation will be presented, and analytical and simulated properties of the model will be discussed. We also present some applications of the model and compare it to more conventional time series models used in hydrology.
H43A-02 1330h
Precipitation Frequency Analysis for China
The Hydrometeorological Design Studies Center of NOAA's National Weather Service recently has made significant changes to their precipitation frequency analysis method and updated their maps for parts of the southwest and east-central U.S. The new method uses statistics called L-moments to select and fit extreme value distributions, as L-moments are less susceptible to outliers in the data. The approach also combines data regionally to determine the shape of the distribution, producing more robust results. In this study, the method is applied to Chinese rain gauge data to create precipitation maps, particularly for 100-year 24-hour precipitation amounts.
H43A-03 1330h
Developing A Geospatial Data Model To Derive Watershed Characteristics For Low Streamflow Prediction
Low streamflow estimates are required for a variety of water resource management purposes. Low streamflow statistics are commonly estimated from a frequency analysis when a historic streamflow record is available, and from a regional regression model when no historic record is available. A regional regression model is developed from low streamflow estimates and watershed characteristics at gauged sites in a region. The developed model is then applied to a hydrologically similar ungauged watershed to estimate low streamflow statistics at that watershed. Watershed characteristics can be derived from digital geospatial data using geoprocessing tools embedded within GIS packages. It has been shown that digitally derived watershed characteristics can lead to model improvements. In order to obtain further improvements on a regional regression model, newly available digital geospatial data (meteorology, topography, geology, etc.) may be used. It is desirable to efficiently process those newly available data. To accomplish this, a common framework and toolset to digitally derive watershed characteristics has been developed in an ArcGIS desktop platform. A relational database has been developed to derive watershed characteristics as a function of attributes of various digital grids. A case study is currently being performed in a region centered on eastern Tennessee and western North Carolina, and includes an assessment of the impact of digital elevation model (DEM) resolution on derived watershed characteristics.
H43A-04 1330h
Estimation of Low Streamflow Statistics at Ungauged Sites Using Baseflow Correlation
Low streamflow estimates are required for water quality and quantity management purposes. This study focuses on estimation of the 7-day 10-year low flow (Q7,10), an extensively employed low flow statistic in the United States. The baseflow correlation method is an information transfer technique that can be used to estimate low flow statistics at an ungauged site by correlating a nominal number of measured baseflows at the ungauged site with those at nearby gauged sites. A national assessment of baseflow correlation estimators is made via a jackknife simulation with daily streamflow values at more than 1300 USGS HCDN gauged river sites. It is shown that the chosen performance metric is important when evaluating the method across a large range of Q7,10 values. Results confirm that baseflow measurements should be obtained from different baseflow recessions. The method performance is sensitive to the correlation coefficient between baseflows at gauged and ungauged sites when the number of baseflow measurements is 5. When the number of baseflow measurements is 10 or more, the method performs adequately if the correlation coefficient is greater than 0.6. The performance of the baseflow correlation method improves as the number of baseflow measurements increases, but levels off dramatically when one has more than 10 measurements. This research also investigates a number of different baseflow correlation methods that employ information from multiple gauged sites to estimate the Q7,10 at a single ungauged site. Results show that the performance can be improved by using multiple site information, especially when less than 10 baseflow measurements are used.
H43A-05 1330h
Simulation of Precipitation at Multiple Stations Using a Multivariate Autoregressive Model With Censored Normal Marginals
Stochastic weather generators are frequently used in climate change studies to simulate input to hydrologic models. In this presentation, we focus on the particular problem of simulating daily precipitation at multiple stations in a region for which records are available. Daily precipitation is a highly intermittent process, highly variable in space, and typically has a highly skewed distribution. A stochastic precipitation model should ideally preserve the regional pattern of intermittence, the autocorrelation, the cross-correlation, and the marginal distributions of observed precipitation. For this purpose, we employed a multivariate autoregressive model. Below zero-values were considered days with no rain. To preserve the marginal distributions of observed precipitation at different stations some prior transformation of data was required. The presentation will describe the experience gained from applying the model to precipitation records in Canada. Focus will be on analytical model properties, methods of parameter estimation, and the preservation of observed statistics in the application.
H43A-06 1330h
Variability at multiple time scales and statistics of hydrologic extremes
The theory of compound distributions has been used recently to investigate the so-called `superstatistics' of complex systems outside of equilibrium. We interpret hydrologic fluctuations as a result of nonlinear dynamical systems driven by hydro-climatic variability at different time scales (e.g., daily and interannual). We explore the role of such fluctuations assuming a clear separation of timescales, and focus on daily precipitation (in terms of both amount and frequency of rainfall) and on the subsequent generation of runoff and streamflow extremes. With the help of simplified stochastic models of rainfall and soil water balance, we discuss how the interactions of such two types of fluctuations can lead, in some conditions, to a scaling behavior in the statistics of extreme hydrologic events such as droughts, intense storms, and river discharge.