ASSESSMENT PREDICTABILITY LIMITS IN SMALL WATERSHEDS TO ENHANCE THE FLASH FLOOD PREDICTION IN WESTERN PUERTO RICO

Transcription

1 ASSESSMENT PREDICTABILITY LIMITS IN SMALL WATERSHEDS TO ENHANCE THE FLASH FLOOD PREDICTION IN WESTERN PUERTO RICO Alejandra Rojas-González Department of Civil Engineering, University of Puerto Rico, Mayaguez PR Eric W. Harmsen Department of Agricultural Eng., University of Puerto Rico, Mayagüez PR Sandra Cruz-Pol Department of Electrical and Computer Engineering, University of Puerto Rico, Mayaguez PR 00680, ABSTRACT Due to the complex terrain and the tropical influence, Puerto Rico is characterized by small watersheds, high rainfall intensity and spatial variability. The rainfall anomalies are produced by orographic-convective type storms, tropical storms and hurricanes producing flash flooding in susceptible areas. An important source of uncertainty in hydrologic modeling in Puerto Rico is associated with the rainfall. There is typically not enough rain gauge density to calculate the associated bias, and to obtain spatial variability of point rainfall at scales below level III radar-based (NEXRAD) products (2x2 kilometers). How the uncertainty in rainfall distribution propagates through the hydrologic modeling process is a critical question, and is addressed in this paper. Another challenge to hydrologic prediction occurs when high slopes exist, and soil and land use characteristics change over short distances. Hydrologic models average the hydrologic parameters and topography in lumped, semi-distributed and distributed models to simplify and/or reduce computational time. The lumping process (i.e., grid upscaling) may lead to a loss in flash flood prediction accuracy. However, it is unknown how much lumping can be tolerated before accuracy of flood prediction degrades beyond an acceptable level. These scientific questions are being addressed within a 4 km x 4 km test-bed watershed, monitored with a network of 28 rain gauges and a stream flow gauge at the watershed outlet. High resolution topography (DEM 10 x 10 meters) and remotely sensed data are being used. The test-bed subwatershed is located in western Puerto Rico and belongs to the Añasco River watershed. The predictability limits in hydrologic response of small watersheds due to grid upscaling methods and rainfall variability are discussed. By quantifying the performance and the prediction limits associated with small subwatersheds, we can understand the behavior of large basins with similar characteristics. This information is needed for developing a real-time flood alarm system in western Puerto Rico, which is the long-term goal of the project. INTRODUCTION A flash flood alarm system developed in a country which is threatened each year by tropical storms and hurricanes is a tool very useful to avoid flood hazards. Due to global warming, tropical storms have increased your recurrence in the hurricanes season. The western Puerto Rico has been flood several times by different storms leaving substantial losses. The requirements to produce an efficient flood prediction are mentioned by the National Research Council and remark that the predictions should require (1) fundamental understanding of the dynamics of the system and propagation of perturbations or uncertainty through it, (2) adequate data to characterize the system states, and (3) procedures for producing the expected evolution of the system. (NRC, 2002). Additionally are necessaries mechanisms for measure the forecast skill and communicate the predictions to users in an effective manner. Real time hydrologic predictions require estimation of stream stage, peak flow, time to peak, and storm volume with high accuracy. To obtain accuracy it is necessary to know and understand the predictability and prediction limitations of the system. Depending on the climatic conditions and physical properties and how these are treated we can obtain different hydrologic responses.

2 In Puerto Rico exist the initiative of develop a real-time flash flood forecasting using innovative radar technology developed by Collaborative Adaptive Sensing of the Atmosphere (CASA) Testbed located in western Puerto Rico. This new radar technology promise revolutionizes the way to detect, monitor and predict rainfall, creating a dense sensor network of low-powered radars that overcome curvature blockage and significantly enhance resolution. The first developed radar was located on Stefanni building at University of Puerto Rico Mayagüez Campus. Few studies have been conducted in Puerto Rico to forecast real-time rainfall runoff. In 1996 the US Geological Survey (USGS) developed a real time rainfall runoff simulation for Carraízo reservoir basin (Sepúlveda et al., 1996). The National Weather Service (NWS) established flood guidance in real time based on soil moisture accounting by regions. Vieux and Vieux (2006) tested a physics-based distributed model in the Loíza basin of Puerto Rico. A long term and event-based simulation were conducted to calibrate the stream flow volume. The soil moisture values calculated in the long term model were fed back into the event-based simulation to enhance the calibration for several individual storm events. A sensitivity analysis to initial soil moisture showed some persistence in antecedent soil conditions, with about one year of warm up. To establish a flood alarm system in Puerto Rico, first it is imperative to know how the watershed behaves under different environmental conditions, parameters spatial variability, input aggregation and associated biases and how these differences are propagated in the solution. This knowledge enhances the forecast skills using distributed models. To understand the system predictability we will conduct various experiments within a small sub-watershed laboratory (test-bed) covering a 4 km x 4 km pixel. This real world laboratory has a rain gauge network with a resolution well below that of the NWS radar products; a stage elevation station at the outlet; high topography resolution (Digital Elevation Model, DEM 10 x 10 meters), remotely sensed data and several field measurements to represent the channel geometry. The test-bed sub-watershed (TBSW) is located in Western Puerto Rico and belongs to the Añasco River watershed. To establish a flood alarm system in the region of the study area, it is necessary to know the performance and the prediction limits associated with the small sub-watersheds, only then we can understand the behavior of larger basins with similar characteristics. The hydrologic based model used in this research is Vflo TM, a fully distributed hydrologic model (Vieux and Vieux, 2002; and Vieux et al. 2003). Vflo TM uses the finite element numerical method to resolve overland and channel flow. The Green Ampt equation is used to represent rainfall infiltration through the soil. The digital revolution in geospatial data has helped to promote the development of physically-based models capable of producing excellent results in flood prediction at internal basin points (Vieux and Vieux, 2002), also this model accept radar rainfall and different forms of distributed rainfall. Study Area Basically, we are developing two hydrologic parallel models, one a big scale that cover around 800 km 2, three major basins and extensive flood plains. The basins comprehended are Río Grande de Añasco, Río Guanajibo and Río Yaguez. In the other hand, the small hydrologic scale model (3.41 km 2 ) where will develop the predictability analysis and will called TBSW. The two models will use radar data from NEXRAD and CASA radars. The Guanajibo, Yagüez and Grande de Añasco rivers have an extensive flood plain and different types of communities are located in these areas. Grande de Añasco River has alluvial fan characteristics which are more difficult for flood prediction, because the accuracy diminishes when the flood area is vast. The greatest flooding in Grande de Añasco River was in September 1998 when Hurricane Georges passed by Puerto. An interval of recurrence greater than 100 years was registered for the Grande de Añasco River. The second largest storm for this watershed occurred in September The Guanajibo River experienced its largest flow during September 1975, 1979 and A lot of events have caused floods in the study area. These events are due to hurricanes, tropical storms or convective rainfall and intense rains. TBSW has around 3.4 km². It belongs to Río Grande de Añasco basin and it is inside of 4 X 4 km pixel. This pixel has a high rain gauge density (26 rain gauges) for CASA radar validation commitment (Figure 1). A high resolution hydrologic model and a predictability study at different terrain and rainfall grid size resolution will be developed. A hydraulic model (HECRAS) was created with GeoRAS and terrain survey to calibrate the rating curve of the creek.

3 Figure 1. Display the three major basin and total area study, (center) Vflo model configuration (200 meters cells) for major study area, (right) TBSW inside the 4x4 km pixel with the 26 rain gages and stage elevation station, and some TBSW photos are displaying the pressure transducer and zone views. Sensitivity Analysis To develop a distributed hydrologic model it is necessary to create an ensemble of different layers that represent the physics characteristics of the basin, also to assign values to the parameter, where some ones are obtained from literature. Uncertainties associated with the model parameters and their scales can be quantified by evaluating the hydrologic response given a range of parameter perturbations. The relative sensitivity coefficient was evaluated with changes of ±50% using the model Vflo, developed by Vieux and Associates, Inc., for 3 events (Sep-1998, Nov-2003 and Sep-2004) and 3 basins; considering the behavior of 2 variables (discharge volume and peak discharge). Results given below indicate that discharge volume was affected principally by the initial soil moisture, followed by overland hydraulic conductivity and soil depth. The peak discharge was affected by roughness, initial soil moisture and hydraulic conductivity. Runoff volume vs. relative changes in parameters is also presented below in the Figure 2. Figure 2. Relative sensitivity coefficient due ±50% in parameter changes (left) and discharge volume changes due to parameter percentage changes (right). Storm Total and Radar Bias Adjustment The Vflo model has the capability to support distributed rainfall and rain gauges data in real time, ideal for a flood alarm system. However, rainfall itself may be the principal source of uncertainty in the model The number of rain gauges in a basin are frequently sparse and therefore do not capture the spatial variability. Two interpolation methods, inverse distance weighted (IDW) and exponential weighted (EW), were compared with radar rainfall from NEXRAD level 3, for November 11-15, It should be noted that the radar is partially dependent on the rain gauge data, as a bias correction factor was applied. (Figure 3). Furthermore, when we using radar it s necessary to remove systematic error applying a correction factor calculated or bias (Vieux, 2004) by the relationship between

4 rain gauges summarize and the radar (Figure 4). For November, 2003 event the bias calculated for whole area was 1.3, the Figure 4 display the scatter plot of radar and raingages and the adjusted line. In the case of September, 2004, Jeanne Storm, the situation it s very particular, because a high reflectivity and rain rate were detected by the radar at southern study area and the rain gages not detected this rainfall amount. The bias calculated was 1.35, but the linear regression has not a slope close to 1. Thereupon, two bias were calculated, one for Guanajibo basin where it was made a 51% reduction and for Río Grande de Añasco basin a bias of 2; obtaining better results. USGS gauge Rainfall (mm) IDW EW Radar Añasco San Sebastian Guanajibo near Hrmigueros Inverse Distance Weighted Exponential Weighted Radar Rainfall (NEXRAD level 3) Figure 3. Total storm display using different interpolation methods and radar data (NEXRAD level 3). The table summarizes the total average rainfall in two stations. Figure 4. Bias correction for Jeanne Total Storm, September 2004, and radar total storm. Bias correction for November 2003 event (right). Slope Calculation Terrain slope is another important source of uncertainty in hydrologic modeling. A method to calculate slope at different grid size resolutions was investigated. Different methods are presented below for the TBSW model. Calculation of the slope from the digital elevation model (DEM10 m x10 m resolution) was determined and then upscaling was performed using four methods. The resample techniques used here were Bilinear, Cubic and Nearest Methods The worst method was found to be up-scaling the DEM and then calculating the slope (red line in the graph). The original slope was more or less preserved and negligible differences were found between the three other techniques (Figure 5). Figure 5. Mean Slope calculated for TBSW using different resample techniques.

5 REFERENCES Abbott, M.B., Bathurst, J.C., Cunge, J.A., O Connell, P.E., Rassmussen, J., 1986b. An introduction to the European Hydrological System Systeme Hydrologique European, SHE, 2. Structure of a physically-based distributed modeling system, J. Hydrol. 87, Beven, K., Binley, A., The future of distributed models: Model calibration and uncertainty prediction, Hydrological Process., 6, Binley, A.M., Beven, K.J., Physically-based modelling of catchment hydrology: a likelihood approach to reducing predictive uncertainty, In: Farmer, D.G., Rycroft, M.J. (Eds.), Computer Modelling in the Environmental Sciences, Clarendon Press, Oxford, pp CASA, CASA Annual Report year 3, Volume II. Engineering Research Center for Collaborative Adaptive Sensing of the Atmosphere, Cooperative Agreement No. EEC , National Science Foundation. March 20, Freer, J., K. Beven, and B. Ambroise, Bayesian estimati on of uncertainty in runoff prediction and the value of data: An application of the GLUE approach, Water Resour. Res., 32, Georgakakos, K.P., Analytical results for operational flash flood guidance, Journal of Hydrology, 317 (2006), Gourley, J. J., Vieux, B. E., A method for evaluating the accuracy of Quantitative Precipitation Estimates from a Hydrologic modeling perspective, American Meteorological Society, April, 2005, McMichael, C. E., Hope, A. S. and Loaiciga, H.A., Distributed hydrological modeling in California semi-arid shrublands: MIKESHE model calibration and uncertainty estimation, Journal of Hydrology, 2005,05,023. Sahho, G.B, Ray, C., De Carlo, E.H., Calibration and validation of a physically distributed model, MIKE SHE, to predict stream flow at high frequency in a flashy mountainous Hawaii stream, Journal of Hydrology, 327, Sepúlveda, N., Pérez-Blair, F., DeLong, L.L., López-Trujillo, D., Real-Time Rainfall-Runoff Model of the Carraízo-Reservoir Basin in Puerto Rico, Water-Resources Investigations Report , U.S. Geological Survey, San Juan, Puerto Rico. Sooroshian, S., Gupta, V.K., Model Clibration, In: Singh, V.P. (Ed.), Computer models of Watershed hydrology, Water Resources Publications, Colorado, USA, pp Urbonas, B. R., Guo, J. C. Y., and Janesekok, M. P., Hyetograph density effects on urban runoff modeling, Proc. Int. Conf. On Comp. Applications in Water Resources, Tamkang Univ., Tamsui, Taiwan, Vieux, B.E., Distributed Hydrologic Modeling Using GIS, Water Science Technology Series, Kluwer Academic Publishers, Norwell, Massachusetts, ISBN , p. 38. Vieux, B. E. Bedient, P. B., Estimation of rainfall for flood prediction from WSR-88D reflectivity: A case study, October 1994, American Meteorological Society, June 1998, pp Vieux, B.E., Bedient, P.B. and Mazroi, E., Real-time urban runoff simulation using radar rainfall and physicsbased distributed modeling for site-specific forecasts, 10 th International Conference on Urban Drainage, Copenhagen, Denmark, August Vieux, B.E., Chen, C., Vieux, J.E., Howard, K.W., Operational Deployment of a Physics-based Distributed Rainfall-runoff Model for Flood Forecasting in Taiwan, IAHS General Assembly at Sapporo, Japan, July 3-11, Vieux, B.E., Moreda, F.G., Ordered physics-based parameter adjustment of a distributed model, In: Duan, Q., Sorooshian, S., Grupta, H.V., Rousseau, A.N., Turcotte, R. (Eds.), Water Science and Application Series, vol. 6., American Geophysical Union, pp , ISBN X (Charpter 20). Vieux, B.E., Vieux, J.E., Evaluation of a physics-based distributed hydrologic model for coastal, island, and inland hydrologic modeling, Coastal Hydrology and Processes, Water Resources Publications, LLC, Highlands Ranch, CO, USA., pp Vieux, B.E., Vieux, J.E., Vflo : A real-time distributed hydrologic model, Proceedings of the Second Federal Interagency Hydrologic Modeling Conference, July 28-August , Las Vegas, NV Abstract and paper in CD-ROM. Wigmosta, M.S., Vail, L.W., Lettenmaier, D.P., A distributed hydrology vegetation model for complex terrain, Water Resour. Res., 30(6),