MEDIUM-RANGE ENSEMBLE FLOOD FORECAST INUNDATION MAPS: THE CASE OF THE TIDAL DELAWARE RIVER NEAR PHILADELPHIA

  • : Ms Word, Ms Word Format
  • : 100 Pages
  • : ₦3000
  • : 1-5 Chapters
  •  
  • Click to DOWNLOAD Materials

MEDIUM-RANGE ENSEMBLE FLOOD FORECAST INUNDATION MAPS: THE CASE OF THE TIDAL DELAWARE RIVER NEAR PHILADELPHIA

ABSTRACT

 

We investigate the ability to enhance flood inundation medium-range (0-7 days) forecast maps through weather ensembles and statistically water surface elevation (WSEL) postprocessing. To generate the flood forecast maps, a one-dimensional hydraulic model (HEC-RAS) is coupled to a regional hydrological ensemble prediction system (RHEPS). The RHEPS is in this case comprised by: i) weather ensembles from the National Center for Environmental Prediction Global Ensemble Forecast System Reforecast version 2 program; ii) distributed hydrological model (HLRDHM); iii) quantile regression (QR) as the statistical postprocessor and iv) verification strategy. The coupled hydrometeorological-hydraulic system is tested in the riverine-estuarine transition zone of the Delaware River near the city of Philadelphia, Pennsylvania. The approach is used to generate 2-hourly high-resolution flood inundation forecast maps at lead times from 0 to 7 days, over the period 2008-2013.

To comprehensively and rigorously verify the forecast maps, the following four different sets of flood maps are generated: i) observed, ii) deterministic, iii) raw ensemble, and iv) postprocessed ensemble. The observed map is generated by forcing the hydraulic model with streamflow and water level observations at the boundary conditions and tributaries of the model. The deterministic and raw ensemble maps use hydrometeorological deterministic and ensemble medium-range forecasts as the forcing, respectively. Given that tide and storm surge forecast were not available for the study period, we force the hydraulic model with observed water levels at the downstream boundary for the deterministic and raw ensemble forecast maps. Lastly, the postprocessed maps are generated by using QR to postprocess the raw ensemble forecasts at individual cross-sections of the hydraulic model.

Results show that the tidal fluctuations of the estuary are highly influential on the forecast skill in the transition zone. Nevertheless, upstream of the head of the tide hydrometeorologic uncertainties are dominant and can cause relatively high errors and biases in the flood inundation forecasts, especially for the later lead times. Moreover, the raw ensembles flood inundation forecasts show higher skill than the deterministic flood inundation forecasts, with higher improvement at lead times from 3 to 7 days. Furthermore, statistical postprocessing improves the skill of the raw ensemble flood inundation forecasts, with an evident improvement across all lead times but is higher at the later lead times. In conclusion, we find that the medium-range flood forecast maps can be skillful and thus provide an alternative for representing and communicating medium-range forecasts. We find that statistical postprocessing can improve the skill of the forecast maps. This may turn out to be a viable approach for bias correcting flood maps in ungauged reaches, which is required for continental scale flood mapping.

 

 

TABLE OF CONTENTS

List of Figures ………………………………………………………………………………………………………….. v

List of Tables …………………………………………………………………………………………………………… vi

Acknowledgements …………………………………………………………………………………………………… vii

Chapter 1 Introduction ………………………………………………………………………………………………. 1

Chapter 2 Study area …………………………………………………………………………………………………. 7

Hydrology of the Delaware River Basin upstream of Trenton …………………………… 7

Hydrodynamics of the tidal Delaware River …………………………………………………… 8

Chapter 3 Datasets ……………………………………………………………………………………………………. 10

Terrain ………………………………………………………………………………………………………. 10

Streamflow ………………………………………………………………………………………………… 11

Precipitation and near-surface temperature …………………………………………………….. 12

Tide  ………………………………………………………………………………………………………….. 14

Chapter 4 Methods ……………………………………………………………………………………………………. 15

Hydrological model …………………………………………………………………………………….. 15

Hydraulic model …………………………………………………………………………………………. 17

Statistical postprocessor ………………………………………………………………………………. 18

Verification strategy ……………………………………………………………………………………. 20

Chapter 5 Results ……………………………………………………………………………………………………… 22

Performance of the coupled model ………………………………………………………………… 22

Performance of the hydrological model …………………………………………………………. 22

Performance of the hydraulic model ……………………………………………………………… 24

Observed map generation …………………………………………………………………………….. 26

Verification of the deterministic flood inundation forecast ………………………………. 27

Verification of raw ensembles flood inundation forecast ………………………………….. 30

Verification of the statistically postprocessed flood inundation forecast …………….. 33

Comparison and illustration of the postprocessed ensemble flood forecast maps

for a selected flood event ………………………………………………………………………. 36 Chapter 6 Summary and discussion …………………………………………………………………………….. 39

Chapter 7 Conclusions ………………………………………………………………………………………………. 42

Bibliography……………………………………………………………………………………………………… 44

Appendix  Metrics for verification ……………………………………………………………………….. 49 Mean Absolute Error …………………………………………………………………………………… 49

Root mean square error ……………………………………………………………………………….. 49

Percent Bias ……………………………………………………………………………………………….. 50

Nash-Sutcliffe Efficiency …………………………………………………………………………….. 50

Wilmott Skill ……………………………………………………………………………………………… 51

Mean Continuous Ranked Probability Skill Score …………………………………………… 51

Chapter 1 

 

Introduction

Floods are among the most lethal of natural disasters in the world, causing over 8 million deaths in the 20th century (Jonkman and Kelman, 2005). In the United States (U.S.), floods are the second deadliest weather related disasters, killing over 4500 people in the

1959-2005 period, showing a high spatial concentration in the eastern United States (Ashley and Ashley, 2008). An operational flood forecasting system with appropriate lead times can help authorities and the public to be better prepared against floods (Cloke and

Pappenberger, 2009). Flood forecasting is usually done by forcing a hydrological model

(spatially lumped or distributed) with numerical weather prediction (NWP) model outputs (Addor et al., 2011; Fan et al., 2014; Pappenberger et al., 2011; Saleh et al., 2016; Siddique and Mejia, 2017). In this approach, the NWP errors and uncertainties are propagated into the flood forecasts. This makes it necessary to explicitly account for uncertainty in flood forecasts.

To account for uncertainty, a widely used method for flood forecasting, in both operational applications and research studies, is the so-called Hydrological Ensemble

Prediction System (HEPS) approach (Addor et al., 2011; Cloke and Pappenberger, 2009; Demeritt et al., 2010; Fan et al., 2014; Khan et al., 2015; Olsson and Lindström, 2008; Thielen et al., 2009a). In the HEPS approach, the weather-related uncertainties are accounted by means of multiple realization or ensembles of atmospheric states produced by the NWP model(s). The weather ensembles are ultimately used to drive the hydrological model and produce a set of possible future streamflow predictions. Although very useful for many applications, HEPS only produce streamflow forecasts (Gouweleeuw et al., 2005; Pappenberger et al., 2011; Saleh et al., 2016; Thielen et al., 2009b). When the goal is to generate flood inundation maps, an additional step must be included in the HEPS to translate the streamflow forecasts into flood maps, where flood extent and depth are spatially shown.

Flood inundation mapping can be done using different approaches (Buahin et al.,

2017; Christian et al., 2013; Di Baldassarre et al., 2010; Leedal et al., 2010; Sarhadi et al.,

2012). The majority of approaches use computational models that solve numerically the Navier-Stokes equations, or a suitable simplification such as the Shallow-Water equations, to produce spatiotemporal representations of hydraulic variables (e.g., water depth, flood extent and water velocity). A common approach to generate flood inundation maps is to couple a hydrological model with a hydraulic or hydrodynamic model (Grimaldi et al., 2013; Lian et al., 2007; Nguyen et al., 2016; Yu et al., 2006). The coupled models can be employed in simulation or forecasting mode. Flood inundation mapping in simulation mode uses a hydrological model, forced with observed meteorological data, to generate streamflow, which are then used as boundary conditions to the hydraulic model. Flood inundation simulation is employed in a number of applications (e.g., flood policy development, flood insurance studies, etc.), and is a topic of current research using both deterministic (Castro-Bolinaga and Diplas, 2014; Costabile et al., 2015; Grimaldi et al., 2013; Ozdemir et al., 2013; Sampson et al., 2015) and probabilistic approaches (Bates et al., 2004; Merz et al., 2007; Sarhadi et al., 2012). Note that flood inundation simulation does not provide information about near-future flood hazards and risks.

In contrast, flood inundation forecasting is done by forcing the hydrological model with forecasted weather variables (usually precipitation and temperature) produced by the NWP model(s). The resultant hydrometeorological forecasts are then used to force the hydraulic model and produce flood inundation forecast maps. Moreover, flood inundation forecast maps can be produced using deterministic or probabilistic approaches (GarcíaPintado et al., 2015; Mashriqui et al., 2014; Nguyen et al., 2016; Pappenberger et al., 2005b; Schumann et al., 2013). Deterministic approaches usually involve the use of more physically based two-dimensional hydraulic model forced by a single deterministic forecast but do not take into account the uncertainty of the input (hydrometeorological) data (Di Baldassarre et al., 2010). While probabilistic approaches to flood inundation mapping use less detailed hydraulic models (e.g., one-dimensional) but account for uncertainty using hydrometeorological ensemble forecasts (Pappenberger et al., 2005b).

There are only a few studies that comprehensively investigate the benefits of using ensemble forecasts to produce flood inundation maps (Buahin et al., 2017; Cooten et al., 2011; Pappenberger et al., 2005b). For example, Buahin et al. (2017) generated probabilistic flood inundation forecasts using the static rating curve (discharge vs stage tables) libraries approach. In this approach, a one-dimensional hydraulic model is used offline to create a library of rating curves by running the model in steady state for a range of flows. Buahin et al. (2017) used the Routing Application for Parallel Computation of

Discharge model to downscale ensemble runoff forecasts from the European Centre for Medium Range Weather Forecasting (ECMWF) general circulation model to obtain local flow forecasts in a basin. Then, they used the rating curve library to convert the flow ensembles into stages. By interpolating stages between cross-sections, they created a flood inundation map for each corresponding member of the ECMWF’s forecast. The static rating curve library approach cannot produce accurate flood inundation forecasts for fully dynamics systems with rapidly changing flows and backwater effects, such as in tidal rivers. Another example of a coupled hydrologic-hydraulic forecasting system is that used by Pappenberger et al. (2005b). They used ECMWF medium-range ensemble precipitation forecasts, for a single storm event, to generate flow ensembles and force a hydraulic model. The focus of their study was in propagating uncertainties through the coupled system rather than in rigorously verifying the system over multiple flood cases.

To improve water level forecasts in tidal and coastal regions, Van Cooten et al.

(2011) developed and employed the NOAA’s Coastal and Inland Flooding Observation

and Warning (CI-FLOW) system. CI-FLOW couples the NOAA’s Hydrology Lab Research Distributed Hydrologic Model (HL-RDHM) (Koren et al., 2004) with the ADCIRC+SWAN model (Dietrich et al., 2010; Westerink et al., 1992), a two-dimensional hydrodynamic model, to generate improved water level forecasts. They applied CI-FLOW to the tidal and neighboring Tar-Pamlico and Neuse rivers, in eastern North Carolina, U.S., using Hurricane Isabel (September 2003) as case study. They forced HL-RDHM with quantitative precipitation forecasts having 5 days of lead time, while ADCIRC+SWAN was forced with pressure wind and wave forecasts from the North American Mesoscale model. From this application, they found good agreement between forecasted and observed water levels at 6 gauging stations (4 tidal and 2 non-tidal) with a better performance in the tidal gages. Van Cooten et al. (2011), however, did not implement ensembles or include extensive verification results.

In this study, we couple a regional hydrological ensemble prediction system

(RHEPS) (Siddique and Mejia, 2017) with a one-dimensional hydraulic model, namely the Hydrologic Engineering Center’s River Analysis System (HEC-RAS), to study the quality of medium-range ensemble water level and flood inundation forecasts in a tidal river. By medium-range, we mean forecast lead times from 0 to 7 days. The RHEPS uses HL-RDHM as the hydrological model and ensemble weather forecasts (precipitation and near-surface temperature) from the National Centers for Environmental Prediction 11-member Global Ensemble Forecast System Reforecast version 2 (GEFSRv2) (Hamill et al., 2013). In addition, to rigorously verify the ensemble forecasts generated with the coupled hydrometeorological-hydraulic system, forecasts are generated (issued) daily at 2-hourly resolution over multiple years (2008-2012). Thus, our main objective is to understand and rigorously verify the skill of ensemble flood forecast maps at the so-called transition zone between the riverine and tidal dominated reach of a river flowing into an estuary region. This transition zone is often a blind spot of research and operational forecasting (Mashriqui et al., 2014) that requires further investigation. In addition, since ensemble flood forecasts generated from weather ensembles tend to exhibit systematic biases (Addor et al., 2011; Siddique and Mejia, 2017), in order to properly verify the forecasts it is necessary to employ statistical postprocessing techniques. Here, we employ quantile regression (QR) to statistically postprocess the ensemble water levels forecasts and thus generate postprocessed ensemble flood forecast maps.

To our knowledge, this is the first study to consider and verify medium-range flood forecast maps in the transition zone, using a coupled hydrometeorological-hydraulic system, weather ensembles, and statistical postprocessing over multiple reforecast years.

Our forecasting approach is demonstrated in the tidal Delaware River near Philadelphia, Pennsylvania, U.S. In this research we investigate whether ensemble hydrometeorological forecasts and statistical postprocessing can improve the accuracy of flood inundation forecast maps in the transition zone. This study is motivated by the following questions: Are weather ensembles able to enhance, over deterministic weather predictions, the mapping of flood inundation forecasts? How skillful are medium-range flood forecast maps in the transition zone? How do the interactions between riverine and tidal flows affect the ensemble flood forecast maps? Can statistical postprocessing improve flood inundation forecast maps? The remainder of the thesis is structured as follows. Chapters 2 and 3 describe the study area and datasets used, respectively. Chapter 4 describes the methods followed to generate, postprocess and verify the ensemble flood forecasts maps. Chapter 5 presents and discusses the main results. Lastly, chapters 6 and 7 contain a short discussion of our findings’ implications and an outline of key conclusions, respectively.

MEDIUM-RANGE ENSEMBLE FLOOD FORECAST INUNDATION MAPS: THE CASE OF THE TIDAL DELAWARE RIVER NEAR PHILADELPHIA

Sharing is caring!

Leave a Reply