Characterizing urbanization impacts on floodplain through integrated land use, hydrologic, and hydraulic modeling

Accepted Manuscript Research papers Characterizing urbanization impacts on floodplain through integrated land use, hydrologic, and hydraulic modeling Avantika Gori, Russell Blessing, Andrew Juan, Samuel Brody, Philip Bedient PII: DOI: Reference:

S0022-1694(18)30824-2 https://doi.org/10.1016/j.jhydrol.2018.10.053 HYDROL 23217

To appear in:

Journal of Hydrology

Received Date: Revised Date: Accepted Date:

19 April 2018 18 September 2018 22 October 2018

Please cite this article as: Gori, A., Blessing, R., Juan, A., Brody, S., Bedient, P., Characterizing urbanization impacts on floodplain through integrated land use, hydrologic, and hydraulic modeling, Journal of Hydrology (2018), doi: https://doi.org/10.1016/j.jhydrol.2018.10.053

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Characterizing urbanization impacts on floodplain through integrated land use, hydrologic, and hydraulic modeling Avantika Gori*1, Russell Blessing2, Andrew Juan1, Samuel Brody2, and Philip Bedient1 Abstract In the United States, fluvial flood risk is managed by the National Flood Insurance Program (NFIP), which delineates areas that have a 1% chance of flooding each year (i.e. the 100-year floodplain). Development within and adjacent to riverine floodplains exacerbates flood losses by increasing peak discharge and runoff volume, shortening the time to peak, and altering the extent of the floodplain. This paper improves understanding of how future development can alter the 100-year floodplain in a watershed in Houston, TX through land use projection, hydrologic and hydraulic modeling, and implements a novel method for considering the impact of both regional trends in development and sitescale development policies on runoff response. The methodology presented in this paper integrates future development scenarios from a machine learning land use projection model with distributed hydrologic modeling and coupled 1D/2D unsteady hydraulic modeling to produce future floodplain estimates. Current site-scale detention requirements are represented within the hydrologic model to evaluate the regional effectiveness of these policies under future development conditions in 2050. Results indicate that the 100-year floodplain can expand by up to 12.5% across the watershed as a result of projected development in 2050 using current stormwater mitigation policies, and the number of parcels within the floodplain can increase by up to 18.8%. This study demonstrates how incremental land use changes can significantly alter the reality of flood

1

Department of Civil and Environmental Engineering, Rice University, 6100 Main St, Houston, TX 77005 Department of Marine Sciences, Texas A&M University at Galveston, 200 Seawolf Parkway, Galveston, TX 77554 * Corresponding author (email: [email protected]) 2

risk in urbanizing watersheds, and how existing land use policies may be insufficient to mitigate impacts from future development. Keywords: floodplain management, flood risk, land use/land cover change, urbanization, distributed hydrologic model 1. Introduction Historical land development trends and policies have greatly exacerbated flood impacts within many urban watersheds across the United States (US) (Leopold, 1968; Ntelekos et al., 2010). This is especially true in watersheds along the US Gulf of Mexico (MichelKerjan et al. 2012), which have experienced high population growth in recent decades (Hallegatte et al., 2013) and are highly vulnerable to flood hazard due to their mild topographic slopes and tropical climates. Fluvial systems in these areas are characterized by slow-draining streams and wide floodplains (Hupp, 2000), and across the country are increasingly impacted by development encroachment into the floodplain (Montz and Gruntfest, 1986; Pinter, 2005). Development within and adjacent to the floodplain exacerbates flood losses by increasing peak discharge and overall runoff volume, shortening the time to peak, and ultimately altering the extent of the floodplain (Hollis, 1975; Rosburg et al., 2017; Suriya and Mudgal, 2012). In the US, fluvial flood risk is characterized and managed through the National Flood Insurance Program (NFIP), which delineates a regulatory 100-year floodplain (the area that has a 1% chance of flooding each year), known as the Special Flood Hazard Area (SFHA). However, there is often a disconnect between the SFHA and observed flood damage (NRC, 2014) that can be attributed in part to outdated floodplain maps, which do not take into account changing land cover impacts on watershed response (Blessing et al.,

2017). For example, one study found that over 60% of floodplain maps were at least 10 years old (Birkland et al., 2003). In high-growth areas like Houston, TX (located along the upper Texas Gulf coast) substantial land development can occur within a span of a few decades that can significantly alter the hydrologic behavior and undermine the SFHA’s capability to accurately represent current flood risk. For example, a recent study found strong correlation between increased urbanization and higher surface runoff in the Cypress Creek watershed in Houston, TX between 2001-2011 (Kim et al., 2016). Thus, there is a need to develop methods that can quantify the impact of incremental land use changes on floodplain extent in order to determine when SFHA maps should be updated. Since NFIP map revisions are often not able to keep pace with changing watershed conditions, development may be occurring in areas that are vulnerable to flooding but have not been designated as SFHAs yet. This paradigm poses a critical challenge to planners and engineers who seek to design long-term flood management strategies in the face of future development uncertainty. In order to accurately quantify evolving riverine flood risk in urbanizing watersheds, it is necessary to develop future land use projections and model their impacts on floodplain extent. Although there have been many studies documenting the increase in peak flows and runoff volumes associated with historical urbanization (Doubleday et al., 2013; Rose and Peters, 2001; Sheng and Wilson, 2009; Vogel et al., 2011), these impacts do not uniformly translate to increases in floodplain extent (Wheater and Evans, 2009). There is limited understanding of floodplain sensitivity to increases in overland runoff rates and volumes, since topographic factors, stream characteristics, and the presence of existing flood infrastructure influence the ability of a watershed to accommodate increases in

overland flow. Although there have been a few studies that have explicitly quantified increases in the 100-year floodplain resulting from urbanization (Habete and Ferreira, 2017; Suriya and Mudgal, 2012), these studies have not considered the effect of existing site-scale development policies, which are intended to mitigate the impacts of new development. There has been little research conducted on the regional effectiveness of site-scale development policies, such as on-site detention/retention, to offset impacts of urbanization. Although there have been some studies examining the local runoff response of these site-scale detention features (Mogollón et al., 2016), there has been no consideration of their efficacy at the regional scale. In addition, to the best of the authors’ knowledge there has been no research examining how well these systems are able to mitigate impacts of future development on floodplain extent. Based on these previous studies two key knowledge gaps emerge: 1) the extent to which future urbanization will increase existing 100-year floodplains and 2) the effectiveness of current site-scale development policies in offsetting floodplain impacts from new development. Previous studies have demonstrated the usefulness of coupling land use projection modeling with hydrologic modeling for watershed response characterization and flood risk quantification. For example, Du et al. (2012) coupled the lumped hydrologic model HEC-HMS with Markov Chain and Cellular Automata land use projection model (CAMarkov) to develop future land use scenarios in the Qinhaui River watershed in China and evaluate their impact on runoff response, finding an increase in daily runoff of up to 13.9% during flood events. Similarly, Choi and Deal (2008) coupled the land use projection model LEAMluc with a semi-distributed hydrologic model (HSPF) to evaluate future urbanization impacts on watershed runoff in the Midwestern US. More recently,

Pumo et al. (2017) implemented a fully-distributed physics-based hydrologic model (tRIBS), a stochastic weather generator, and a CA-Markov land use model to characterize the joint impacts of climate and urbanization on watershed hydrologic processes. Wijesekara et al. (2012) also implemented a fully-distributed model to project changes in streamflow in a Canadian catchment until 2031. However, there has been no implementation of this coupling methodology for floodplain evolution assessment. These previous studies have primarily investigated annual or daily runoff impacts of urbanization, without considering the impact that these changes may have on flood inundation potential. This study aims to address the limited understanding of watershed sensitivity to future development and the regional impacts of site-scale mitigation by utilizing an approach that integrates land use projection modeling with hydrologic and hydraulic modeling to quantify floodplain increases under a range of future development conditions. By linking land use projection modeling with hydrologic/hydraulic analysis, this paper evaluates the impacts of both regional patterns of development as well as site-scale mitigation actions, effectively considering the crucial feedback loop between anthropogenic activities, hydrologic response, and natural hazard management. First, a land use projection model for the region is developed and used to predict a low and high growth scenario for the year 2050, a time horizon chosen because it generated enough new development to have significant hydrologic impacts. These scenarios are represented within a distributed hydrologic model to evaluate the impact of urbanization patterns as well as site-scale development requirements on 100-year overland flows. Flow hydrographs are linked to an unsteady hydraulic model to assess development impacts on

water surface elevations, and ultimately produce current and future 100-year floodplain depths and extents. Floodplain results are analyzed to characterize watershed response to future development, identify potentially vulnerable regions of the watershed, and provide information about the effectiveness of existing policies in mitigating future flood risk. This approach is applied to a case study watershed in northwest Harris County that has experienced rapid development in recent years, has been highly vulnerable to riverine flooding, and is expected to continue developing rapidly in the future. The following sections outline the study area and provide an overview of the modeling approach. This is followed by detailed methods for each modeling step: land use projection, hydrologic analysis, and hydraulic analysis. Case study results are presented for current and future development conditions with increases in flood risk measured as increases in floodplain extent and inundated parcels. These results are discussed within the context of regional flood management planning and local development policy. This paper concludes with a summary of the value of this approach for regional planning purposes and provides recommendations for future work. 2. Materials and Methods 2.1 Study Area The Cypress Creek watershed is located north of the city of Houston in northwest Harris County, along the upper Texas Gulf Coast (figure 1). This watershed serves as a prime case study area to apply this modeling approach because it has experienced high rates of development in recent years and recent precipitation events have highlighted the severe flood vulnerability of residents within the watershed. The watershed encompasses a 692 km2 drainage area, features over 400 km of open drainage channels, and contains a

population of 347,334 (HCFCD, 2017). The elevation of the watershed ranges from 6.6 m to 95.4 m above mean sea level (MSL), and features an average overland slope value of 0.51%. Although the northwestern portion of the watershed has steeper slopes and the southern portion has milder slopes, the vast majority of overland slopes range from 0% to 2%. The Cypress Creek watershed is currently partially developed, with the majority of developed land located on the east side of the watershed. The western portion of the watershed is primarily composed of agricultural and natural prairie land (figure 1). Cypress Creek serves as the primary drainage conduit for the watershed, draining west to east until flowing into the San Jacinto River. The stream is slow-draining since it has largely remained in its natural state, with vegetation lining the banks and natural meanders. However, due to the flat topographic slopes in the southwestern portion of the watershed and the limited conveyance capacity of the stream, water spills over the watershed divide and into the neighboring Addicks Reservoir watershed during highintensity rain events (HCFCD, 2017). Inter-basin overflow can occur during rain events greater than a 5-yr magnitude (14.7 cm in 24hr), and result in significant volumes of overflow during higher intensity events (HCFCD, 2017). This complex hydraulic phenomenon is somewhat unique and poses a challenge for modeling the rainfall-runoff response of the watershed, since overflow rates depend on both rainfall intensity and downstream water surface elevations. Thus, traditional 1D models have been unable to simulate the runoff dynamics during extreme rain events and are unable to accurately represent the 100-year floodplain in this portion of the watershed. The Cypress Creek watershed has experienced several major flooding events over the last few years that have inundated thousands of homes and resulted in substantial economic

losses. During a storm event in April 2016 known as the Tax Day flood, for example, over 2000 homes were flooded in the Cypress Creek watershed alone, and peak water elevations throughout the watershed far exceeded previous records (Lindner and Fitzgerald, 2016). More recently, during Hurricane Harvey, thousands of residents in the watershed experienced severe inundation over a period of several days (Sebastian et al., 2017), leaving hundreds stranded in their homes. Recent flood events in the Cypress Creek watershed are even more concerning when considering the rapid development simultaneously occurring in the area. From 2000 to 2010, the population of the watershed grew by 70% on average, while development rates in the western portion were as high as 390% in one zip code (Zheng, 2011). 2.2 Methodological Approach Overview The methodology presented in this paper links land use projection modeling with hydrologic and hydraulic modeling to more accurately understand the potential impacts of a range of future development scenarios on the delineation of the 100-year floodplain. This approach consists of three main components: 1) land use projection modeling to produce future development scenarios; 2) hydrologic modeling to quantify changes in basin and tributary discharge rates; and 3) hydraulic modeling to assess increases in maximum water surface elevation and map changes in 100-year floodplain extent (figure 2). Future land use scenarios in the Cypress Creek watershed are developed for 2050 in the form of a low development scenario and a high development scenario. These scenarios serve as input to a calibrated hydrologic model, and their impacts on overland flow are modeled for a 100-year design storm. Discharge hydrographs produced within the hydrologic model are linked to a coupled 1D/2D unsteady hydraulic model. This

model routes discharge hydrographs through the channel to produce stage hydrographs along Cypress Creek and ultimately quantify the 100-year maximum water surface elevations. Finally, the maximum water surface elevation is used to map the extent and depth of the 100-year floodplain. The following sections describe the methodology of each component in the modeling approach in more detail. 2.3 Land Use Projection Model In order to assess the impact of future development in the Cypress Creek watershed, future land use scenarios for the year 2050 were developed based on historical urbanization patterns in the Houston-Galveston region. Although there is considerable uncertainty introduced by utilizing historical development patterns to project future growth in 2050, this approach is reasonable when considering the unique characteristics of the Houston region. The city has largely expanded radially from the downtown center over the past 50 years, mostly due to the lack of zoning regulations (Qian, 2010), making it a relatively straightforward region to project development in the future. Land use projections developed in this study represent a “business-as-usual” situation, where future development occurs in largely the same way it has occurred over the past half-century. A multi-layer perceptron artificial neural network (MLPNN), which is a widely used pattern recognition model (Pijanowski et al., 2002) was selected to predict where development is likely to occur in 2050. This model was selected because it has been shown to consistently outperform other commonly used land-use change models (Lin et al., 2011). Artificial neural networks (ANNs) are particularly suitable for quantifying the relationships between urbanization and drivers of development because they do not make assumptions about the normality in datasets, can capture nonlinear relationships, and can

account for complex interactions between driver variables (Ji, 2000). The MLPNN developed in this study utilizes a multivariate function that predicts the probability of transitioning from undeveloped to developed at any location based on the values from the driver variables.

2.3.1 Model Setup and Scenario Development A variety of datasets were used in this study to model urban growth. Land use/land cover (LULC) maps were obtained from the National Land Cover Dataset (NLCD) for the time periods 2001, 2006, and 2011 at a 30 m resolution. These land cover datasets were selected because of their accuracy, direct comparison capability across time, and because they capture a decade of LULC change (Homer et al., 2015). Based on the focus of this study (urban growth) and the dominant landscape features of the study area, the land cover imagery was reclassified into six LULC classes, which included developed, barren, forest, grassland, agriculture, and wetlands. The remaining datasets were used to develop the a priori drivers of urban growth. However, there is no universal set of drivers of urban growth, instead drivers were selected based off of findings from previous studies (Arsanjani et al., 2012; Hu and Lo, 2007; Shafizadeh Moghadam and Helbich, 2013; Thapa and Murayama, 2012) and available data within the study area. Sources of data for the development of urban growth drivers included the Census Bureau, the National Hydrography Dataset (NHD), Houston-Galveston Area Council (HGAC), and Texas Education Agency (TEA). The drivers derived from the datasets were grouped into three categories: biophysical, infrastructural, and socioeconomic (Table 1). Euclidean distances were used to measure distance to coast, streams, developed, downtown, schools, roads,

and parks. The two categorical variables (land cover type and Census place) were converted into continuous variables using the evidence likelihood transformation, which calculates the relative frequency of pixels that were developed between 2001 and 2006 within each category and then assigns each category that value. Finally, percent economically disadvantaged was used as a proxy for school district performance as found in a recent national study (Rothwell, 2012). A critical component of predicting future urban growth is the selection of the drivers that are associated with regional patterns of LULC change. The selection of drivers was done in two parts: the first was a preliminary exploratory analysis of the association between the a priori drivers and historical changes in development, and the second was a more rigorous backwards stepwise constant forcing procedure. In the first part, Cramer’s V was used to test the potential explanatory power of each driver layer and development that occurred between 2001 and 2006. Cramer’s V ranges from 0 to 1, where a high value would indicate that a driver has a strong association with urbanization and a very low value would indicate that the driver should likely be discarded. For this analysis, drivers were removed if they below a Cramer’s V value of 0.15 (Eastman, 2012). The second part (i.e. backwards stepwise constant forcing) is conducted after the MLPNN transition model is finished running with all the remaining drivers. This procedure examines the relative power of each driver variable and then iteratively removes drivers starting with those that have the least power until the skill of the model begins to decline. This procedure not only removes drivers with relatively little power in determining where development will occur, but also reduces the likelihood of overfitting.

Transition potential modeling begins once the final set of drivers has been selected. At this stage, an MLPNN is trained on a sub-sample of cells within the study area, half of which transitioned to development and half of which remained the same from 2001 and 2006. To train, the MLPNN constructs a network of neurons between the driver variables and the types of transitions that occurred. The MLPNN iteratively adjusts the weights between neurons until an optimal solution is found (i.e. minimizes the root mean square error and maximizes accuracy). The MLPNN produces an accuracy rate which expresses the capacity of the model to find the best combination of driving factors based on selflearning and self-modification. An accuracy rate within the vicinity of 80% is considered acceptable (Eastman, 2006). Model validation was tested using the area under the receiver operating characteristic (AUC), which has been efficient at evaluating accuracy of LULC change models (Peterson et al., 2008). AUC is based on an examination of correctly and incorrectly predicted change and generates a curve termed the receiver operating characteristic (ROC) curve. AUC ranges from 0 to 1, where 0.5 represents randomness or no skill and a value of 1 represents perfect skill (Sangermano et al., 2010). This test was conducted by comparing a modeled transition potential map for the year 2011 with what was actually developed in 2011. The transition potential map is essentially a 2011 development suitability map that calculates the probability of transitioning from undeveloped to developed for each cell. Transitions were modeled using the MLPNN with its associated underlying driver variables. Finally, future urban growth demand was modeled using a Markov chain procedure, which determined the amount of development expected to occur in 2050 based on the conditional probability of each land cover type transitioning to development and historic

rates of change. Constraints to development (e.g. parks, conservation easements, and water bodies) were incorporated to prevent these areas from being erroneously developed. The model produced two outputs: 1) a probability map for the entire watershed, depicting the probability each cell would become developed by 2050, and 2) a predicted development map, which represented the model’s “best guess” at which areas would become urbanized by 2050. These outputs were then used to derive two scenarios: high and low growth. The scenarios were derived by expanding the 2050 prediction by 15% for high growth and shrinking the 2050 prediction by 15% for low growth based on the “best guess” scenario plus/minus 15% more area based on the probability map. These scenarios capture much of the uncertainty within the model and also produce a range in development that could occur given changes in development trends and land use policies. 2.3.2 Model Performance Results Much of the built-up area within the Cypress Creek watershed in 2001 was located in the far eastern part of the watershed closest to downtown Houston. Between 2001 and 2006 development expanded westward converting approximately 59.32 km2 of undeveloped land most of which was forest (46%) and agricultural land (31%). All the drivers except two (i.e. distance to parks and Census place) had an acceptable association (Cramer’s V value > 0.15) with development. Multiple iterations of backwards stepwise constant forcing removed all the drivers that did not affect model performance. The result of this process eliminated all the drivers except land cover type, distance to development, distance to downtown, and distance to schools. The final model with the four remaining drivers had an acceptable MLPNN accuracy rate of 82.48% and actually performed better than the model with all of the drivers (i.e. accuracy rate increased by 1.76%). This is

likely a result of overfitting in the initial model due to the presence of too many variables with relatively poor explanatory power. Finally, AUC was used to measure the prediction accuracy of the MLPNN by comparing the transition probability prediction map of 2011 against what was actually developed in 2011 which resulted in an AUC of 0.809, which is considered acceptable (Shooshtari and Gholamalifard, 2015). 2.4 Hydrologic Model This study utilizes Vflo®, a physics-based, distributed hydrologic model, to simulate the rainfall-runoff process. Vflo® solves conservation of mass and momentum equations using a finite-element approach, and represents the physical characteristics of a watershed in gridded-cell format (Vieux and Bedient, 2004). In the model domain each grid cell contains parameters that represent elevation, soil type, land cover characteristics, and a flow direction that is defined based on relative elevation compared to surrounding cells. Grid cells can be designated as overland or channel cells, and channel cross sections can be extracted from digital elevation data. The model performs rainfall-runoff calcaulations within each grid cell, and overland flow between cells is routed via the Kinematic Wave Analogy (KWA), which is a simplification of the 1D Saint-Venant equations. A full description of the KWA derivation is documented in (Vieux & Vieux, 2002). Infiltration is calculated at each grid cell using the Green & Ampt Equation, which depends on soil parameters of hydraulic conductivity, wetting front suction head, effective porosity, and soil depth (Rawls et al., 1983). In this study, Modified Puls routing is utilized to model channel flow for the tributaries of Cypress Creek, since it is more suitable for representing channel storage in mild-sloped watersheds (Vieux and Bedient, 2004).

Distributed models are particularly useful for representing spatially-diverse land cover characteristics and modeling land cover evolution through time since calculations are made at the grid cell level. In contrast, traditional lumped modeling methods often rely on empirical parameters derived at a subbasin-scale, which may not be able to accurately represent localized development changes (Blessing et al., 2017). Vflo® has been successfully utilized to model development scenarios, low impact development features, and flood mitigation infrastructure (Doubleday et al., 2013; Fang et al., 2010; Juan et al., 2017). Additionally, Vflo® was chosen because it has been widely applied and validated in the Houston region for both inland and coastal watersheds (Blessing et al., 2017; Ray et al., 2011; Vieux and Bedient, 2004). 2.4.1 Model Setup and Calibration/Validation Vflo® model setup requires detailed elevation, soil type, and land cover information. The model domain was delineated based on the Harris County Flood Control District (HCFCD) watershed boundary and utilized a grid cell resolution of 91 m (300 ft), which was determined based on a maximum model size of roughly 100,000 cells. 2008 LiDAR Digital elevation (DEM) data was obtained from the Houston-Galveston Area Council (HGAC), and was utilized in the model to determine the slope of each cell and the overland flow direction grid, and to extract cross-section profiles for channel cells in the model. Soil type information was obtained the Texas Natural Resources Information System (TNRIS), and processed in ArcGIS according to Rawls et al. (1983) to obtain estimates of hydraulic conductivity, effective porosity, wetting front capillary pressure head, and soil depth.

Land cover data to represent current conditions in the watershed was obtained at 30 m resolution from a 2011 dataset within the National Land Cover Database (NLCD). Vflo® is able to represent land cover/use through a Manning’s roughness coefficient and a percent imperviousness applied at each grid cell. Roughness coefficients indicate the amount of frictional losses between flowing water and ground surface, and impact the veloctiy of overland flow (Kalyanapu et al., 2009). Consequently, natural areas of high vegetation or forest will have higher roughness coefficients and lower flow rates, while concrete or pavement areas will have low roughness and high flow rate. Land cover categories from NLCD are converted to Manning’s roughness coefficients based on Kalyanapu et al (2009), and impervious percentages are designated based on NLCD guidelines (NLCD, 2011).

The hydrologic model was calibrated using two significant rainfall events, one on April 17th 2016 and the other occurring on May 26th 2016. Rainfall data for both events comes from Next Generation Weather Radar (NEXRAD) Level II, and was obtained at 4 km spatial resolution and at 5 minute intervals. Radar data was then calibrated against local rain gauges before applying to the Vflo® model. The first rain event in April 2016 was an extreme precipitation event that dropped over 38 cm in 12 hrs on some parts of the watershed, and resulted in the Tax Day flood described in section 2.1 (Lindner and Fitzgerald, 2016). This storm exceeded a 500-yr frequency event in the western portion of the watershed and a 100-yr frequency on average throughout the study area (Lindner and Fitzgerald, 2016). The second storm occurring on May 26th 2016 resulted in 12.717.8 cm across the watershed, corresponding to roughly a 10-yr frequency event. These

two events were chosen because they represent a range of frequency storm magnitudes, and because they occurred during the same time period. This latter point is important in order to isolate and calibrate to the most recent development conditions. Calibration procedure for the model focused on adjusting soil parameters of hydraulic conductivity and soil depth, which control the rate of infiltration and available ground storage. Additionally, channel roughness values were adjusted to fine-tune the timing and peak of the rising limb of the stream hydrographs. Calibration efforts focused on capturing the rising limb and peak of the observed hydrographs, since these are important characteristics for flood modeling purposes. The initial soil saturation parameter within the model was adjusted for each calibration event based on the presence of rainfall within the previous two weeks. Since no significant rainfall was observed before the April 2016 storm, the soil saturation to set to zero. For the May 2016 event, there were several centimeteres of rainfall that occurred one week before the event, so the soil saturation was set to 20%. Five USGS streamflow gages along Cypress Creek were used as calibration points (figure 3a). The average peak flow difference and Nash-Sutcliffe Efficiency across the five gages was 2.1% and 0.80, and -1.2% and 0.74, for the April 2016 and May 2016 storms, respectively (with negative values indicating under-prediction by the Vflo® model). Based on these performance metrics and the overall shape and timing of the comparison hydrographs, the authors believe these are satisfactory calibration results. Figure 3a shows hydrograph comparisons at the middle gage location, which is the most reliable gauge (based on rating curve measurements) in the watershed, and figure 3b shows a comparison of modeled peak streamflow vs observed. Although the falling limb of the

hydrograph is not as well-captured by the Vflo® model for the April 2016 event (shown in figure 3a), the rising limb and peak match well with observed, demonstrating the model’s ability to adequately capture the flood wave. In addition to calibration, the Vflo® model was also validated against two additional rainfall events; one was a smaller event (15.1 cm) that occurred on March 15th 2016 and the other was Hurricane Harvey’s extreme rainfall (75.4 cm), which occurred August 25th-30th 2017. Rainfall data for these events was also from NEXRAD Level II radar at 4 km resolution. The average peak flow difference across the five gages was -2.4% and 8.22% for the March 2016 and August 2017 storms, respectively (with negative values indicating model underprediction). Figure 3c shows the modeled vs observed peak flows for these two storms, indicating acceptable performance for both events. For the August 2017 event modeled results matched well with observed for all gages except gage #4, where the model predicted a significantly higher peak than the observed. However, it is possible that this gage was malfunctioning during the storm, since the match at the next downstream gage (#5) was good (7.1% difference in peak flow). In order to ensure accurate comparison between current and future conditions, a 100-year design storm was applied to the current conditions Vflo® model to generate 100-year flow hydrographs. The 100-year rainfall hyetograph for this region corrseponds to 31.5 cm in 24 hrs, and was applied according to the Soil and Water Conservation Society (SWCS) guidelines for a Type-III 24 hr storm event, which is the same rainfall methodology applied by Harris County floodplain managers in modeling a 100-year event (Storey et al., 2010). 2.4.2 Site-Scale Detention Modeling Methodology

Although roughness coefficients from Kaylanapu et al. (2009) are used to represent frictional losses for current development in the watershed, these values were not applied to represent future development. These values assume no on-site detention measures, and thus represent development impacts under a no mitigation scenario. Instead, this study derives new roughness values based on development criteria from the HCFCD (which is the primary floodplain regulatory agency in Harris County), and the topographic conditions of the Cypress Creek watershed. According to HCFCD’s Policy, Criteria, and Procedure Manual (Storey et al., 2010), new developments must adhere to peak discharge rate restrictions that are defined for a 10year and 100-year storm event. For developments smaller than 2.59 km2, a simple Site Runoff Curve can be used to determine the maximum allowable peak discharge. For developments larger than 2.59 km2, more rigorous hydrologic and hydraulic modeling is needed to determine the appropriate on-site detention requirements. As a point of reference, the size of each grid cell in the Vflo® model is 0.008 km2. For the purposes of this study it was assumed that all new developments would be the size of a single Vflo® grid cell, so on-site detention could be modeled using the Site Runoff Curve equations to constrain the maximum discharge rate. These curves are determined based on the following equation: 𝑄 = 𝑏𝐴𝑚

(1)

Where Q is peak flow rate, A is development area, b is a factor based on impervious percent of the development, and m a factor based on the size of the development. For development areas less than 0.08 km2, m is equal to one and the equation simplifies to a linear relationship. Using this equation, each grid cell in the Vflo® model that is

projected to become developed in 2050 should have a peak discharge of less than 0.14 m3/s. This methodology for calculating and applying the maximum discharge rate to each individual cell is appropriate for small/medium developments because given the linear nature of the Site Runoff Curves, a group of developed cells would have a combined peak discharge that is still in compliance with the detention requirements for their collective area. For example, a new development the size of ten Vflo® grid cells has a maximum allowable discharge ten times that of an individual developed cell. However, for larger developments greater than 0.08 km2, the m parameter changes to 0.823, which results in smaller peak flows than would be calculated by using m equal to one. Since this study only utilizes the approach for small/medium developments, it may over-predict the flow impacts, since in reality the watershed might feature large developments as well that have a lower peak flow per acre rate. Still, this approach allows site-scale policies to be approximated at the watershed-scale to facilitate more accurate analysis of flood impacts, and by assuming only small/medium developments represents a more conservative (worst case) scenario for future development impacts. On-site detention features were modeled at the grid cell level by adjusting the overland roughness parameter, which impacts the rate of flow from the cell. Additionally, the slope value of each grid cell also impacts the peak flow from that cell. Since each cell across the watershed has a different slope value, it is necessary to develop an appropriate roughness parameter that can ensure cells with varying slope values are all still in compliance with policy guidelines. Thus, a characteristic slope value (so) for the entire watershed was defined based on the distribution of slopes. Since steeper slopes produce higher peak flow values, the 90th percentile value of the slope distribution was chosen to

ensure that all slope values less than this threshold would be in compliance with the peak flow restrictions (i.e. Q in equation 1). S0 is applied to a single design grid cell and a 100year rainfall is applied on this cell. The overland roughness parameter is iteratively adjusted until the peak flow from the design cell reaches Q = 0.14 m3/s (as calculated from equation 1). This adjusted roughness value is defined as r0, and represents the increased roughness from the presence of on-site detention features. Next, r0 is applied to all newly developed cells, and the 100-year peak flow from a random sample of 100 cells is computed to ensure that the vast majority of cells are in compliance using this updated roughness value, and that no significant approximation errors are introduced. It should be noted that s0 is only utilized to determine an appropriate roughness value (r0), and the analysis is performed using the actual slope value of each grid cell. This process is summarized in four main steps in figure 4. Based on the distribution of slope values (figure 4 step 2), the characteristic slope value (s0) was determined to be 2.0%. Then by applying this 2.0% slope to a single Vflo® cell and evaluating the peak flow from a 100-yr rainfall (figure 4 step 3), the new roughness value (r0) was determined to be 0.2, compared to 0.0678 under a no mitigation scenario (Kalyanapu et al, 2009). This representative roughness was applied to all newly developed cells within the watershed and the 100-year peak flow for a random sample of 100 cells was tested to ensure compliance across the watershed (figure 4 step 4). Of the 100 cells tested, 89 complied with peak flow requirements, and only 11 exceeded the peak flow threshold. However, most of these cells were within 5% of the required peak flow and all cells were within 10% of the requirement. Thus, this performance was deemed acceptable for modeling site-scale detention.

2.5 Hydraulic Model For this project, hydraulic analysis was conducted using a hydraulic model, HEC-RAS (Hydrologic Engineering Center – River Analysis System), developed by the U.S. Army Corps of Engineers. This model has been used in a wide variety of applications, including floodplain assessment, flood insurance studies, and dam breach analysis (Bass et al., 2017; Butt et al., 2013; Knebl et al., 2005; McLin et al., 2001). The primary function of HEC-RAS is to calculate water surface elevations at channel cross sections or modeled storage areas of interest for any given flow rate. The latest version of the software, Version 5.0.3 (HEC, 2016), is capable of computing water surface profiles by performing one-dimensional (1D), two-dimensional (2D), or combined 1D/2D hydraulic calculations, based on energy and momentum equations. In this study, the effective hydraulic model of Cypress Creek watershed developed by HCFCD was used as reference. This model is a 1D-steady model, and is used as the basis for generating the 100-yr FEMA floodplain. The model generates a static water surface profile (i.e., maximum water surface elevation) along the entire channel based on peak discharges inputted at specific channel cross sections. While a 1D-steady model is useful for floodplain assessment and floodway encroachment studies, it is insufficient to model the hydraulic performances of intricate systems where volume and timing are crucial, such as the Cypress Creek overflow area. In order to simulate the hydrodynamics at this particular location, the effective HCFCD model was modified and converted to a 1D/2D unsteady hydraulic model. The main advantage of an unsteady model compared to a steady model is that it can simulate the water surface profiles of entire storm hydrographs instead of just peak flows, which provides a better understanding of the system’s flow

and stage response over time. This is crucial for modeling the overflow area in Cypress Creek, because the overflow dynamics depend on both stage hydrograph timing and peak. The HEC-RAS model was validated using the same two precipitation events that were used to calibrate the hydrologic model and additionally validated against the August 2017 event (for a total of three validation events). For each storm event, basin and tributary flow hydrographs from the hydrologic model served as boundary data to the HEC-RAS model at specified cross sections. Observed stage hydrographs for each storm served as the downstream boundary condition for the HEC-RAS model, since a stage gauge was located close to the outlet of Cypress Creek (figure 5). Finally, initial water levels were determined by setting base flow as 1% of peak flow for each storm event. Since base flow in Cypress Creek is negligible during normal conditions (the creek is only fed by rainwater), these initial conditions were implemented to improve model stability, as recommended by HEC (2016). Validation locations were determined based on the presence of high water marks in addition to peak water levels recorded at gauges along Cypress Creek. Maximum water depths modeled in HEC-RAS were compared to high water marks along Cypress Creek for the April 2016, May 2016, and August 2017 storms. Additionally, for the April 2016 and August 2017 storm events there were several high water marks recorded in the overflow area, which were used to ensure a good match between modeled overflow depths and observed depths. The average peak depth difference was -0.01 m, -0.3 m, and -0.12 m for the April 2016, May 2016, and August 2017 storms, respectively (with negative values indicating model under-prediction) and a comparison between modeled water depth and observed depth for both storms is shown in figure 5. Although the model

slightly under-predicted depth for the May 2016 event, it produced good results for the April 2016 event, which is close to a 100-year magnitude, and for the August 2017 event, which was greater than a 500-year event. 3 Results 3.1 Future Development Scenarios As shown in figure 6, the Cypress Creek watershed is currently partially developed, with a majority of the watershed composed of agricultural and natural lands. Development projections for 2050 predict that new development will occur in areas adjacent to existing development, such as the undeveloped lands located in the middle and eastern portion of the watershed, and new development trends will push westward into areas currently dominated by natural and agricultural land cover. A comparison of 2011 land use, 2050 low development, and 2050 high development shown in figure 6 illustrates the trend of westward development, and shows that the primary difference between the low and high development scenarios is the amount of development located in the western portion of the watershed. Table 2 indicates that development in 2050 is projected to grow by 37%-54.5%, becoming the dominant land use type within the watershed. While developed land sees the largest gains in 2050, natural lands are projected to experience the highest losses, decreasing by 54%-61%. Agricultural lands remain relatively constant, shrinking by only 11%-24%. These results suggest that future development within the Cypress Creek watershed could disproportionately impact natural land, such as forests and wetlands, compared to pasture and crop land. Figure 6 demonstrates this trend further, showing that large areas of natural land are projected to become developed in 2050.

3.2 Floodplain Extent Figure 7 illustrates the extent of the 100-year floodplain under each of the three scenarios: 2011 conditions, 2050 low development, and 2050 high development. A comparison of the three floodplains illustrates that increases in floodplain extent are moderate along the middle and downstream portion of the watershed, but are more severe in the upstream portion, which is magnified in figure 7A. This magnified area corresponds to the inter-basin overflow area, where Cypress Creek spills over its banks and into the neighboring Addicks Reservoir watershed. Figure 7B shows a portion of the midstream of Cypress Creek. In this region, floodplain increases are much less severe than in the overflow area. Table 3 shows the change in 100-year floodplain extent and increase in inundated residential parcels between 2050 development scenarios and 2011 conditions. The 100year floodplain extent is projected to increase by 8.4-12.5% across the watershed, which corresponds to an increase in inundated area of 9-13 km2. The impact to residential parcels is more severe, ranging from 12.3-18.8% increase compared to current conditions. Across the watershed, this corresponds to an additional 361-550 impacted parcels. These estimates only include existing residential parcels, and do not take into account newly developed parcels in 2050. Thus, it is likely that Table 2 under-estimates the true increase in residential flood risk across the watershed. Instead, it represents existing parcels that could become designated as special flood hazard areas (SFHAs) in the future due to growth of the floodplain extent. Table 4 displays similar flood risk statistics as Table 3 for the overflow area alone. Within the overflow area, the floodplain extent is projected to increase by 16.4%-23%,

and increase the number of inundated parcels by 21%-25%. These impacts are considerably greater than results across the entire watershed. Figure 8 depicts the channel morphology and increase in maximum water level (in meters) for two different cross sections in Cypress Creek: one located in the upstream overflow area (inside box A in figure 7), and one located in the midstream portion of the watershed (inside box B in figure 7). Along the overflow area the increase in maximum water level ranges from 0.05 m to 0.07 m, corresponding to a low and high development scenario, respectively. In contrast, the midstream area sees a 0.2 m to 0.34 m increase in maximum water level. Figure 8 also highlights the differences in channel morphology between the two areas, depicting the flat slopes and limited channel storage present in the overflow area, and the relatively steeper slopes and high storage capacity in the midstream area. Although the ground elevation (relative to MSL) is different in each location, figure 8 shows the same elevation change (10 m) between the two cross sections. 4 Discussion 4.1 Watershed Sensitivity to Future Development The results indicate that the 100-year floodplain can expand by nearly a quarter of its original size as a result of nearly four decades of projected urbanization in the Cypress Creek watershed. In general, across the watershed, a 1% increase in development translated into about a 0.23% increase in the extent of the floodplain. The overflow area in particular was much more sensitive to future urbanization, with impacts two times more severe than across the entire watershed. Even under a low development scenario, the floodplain is projected to increase by more than 20% in the overflow area, indicating

that this region is highly sensitive to urbanization impacts. In contrast, the middle and downstream portion of the watershed appear to be fairly resilient to increases in development, since they are able to accommodate large increases in future development without substantial increases in floodplain extent. The difference in floodplain impacts between the overflow area and the middle/downstream portions of the watershed is largely determined by changes in slope and channel storage capacity. In the overflow region, the ground slope is mild and there is little storage capacity outside the channel, as shown in figure 8. This results in a wide floodplain extent even under current conditions, since relatively small increases in water elevation result in large increases in inundation extent. Furthermore, since Cypress Creek already has limited storage capacity in this location (described in section 2), increases in runoff volume resulting from new development cannot be effectively stored or conveyed through the channel. Instead, excess runoff volume spills over the channel banks and drains through the overflow area, resulting in substantial increases to the 100-year floodplain in this area. In contrast, the middle and downstream portion of the channel has a large amount of overbank storage, which is able to constrain the floodplain extent. Previous studies have also shown that the presence of overbank storage capacity can improve flood wave attenuation by storing excess floodwater (Castellarin et al., 2011; Woltemade and Potter, 1994). Although the middle and downstream portions of the watershed do not experience a significant increase in floodplain extent, these areas are still impacted by excess runoff from new development. The midstream portion of the watershed experienced the largest increase in water depth, up to 0.34 m under a high development scenario. In contrast, the

overflow area experiences only a small increase in water level under both development scenarios. This difference in water level increase between the two areas suggests that in the downstream region the effects of urbanization are translated into increases in flood depth rather than flood extent due to the topographic conditions of the area. Consequently, although floodplain expansion is constrained in the middle and downstream portions of watershed - which limits the increase in the number of newly inundated parcels - parcels that are already within the 100-year floodplain in these locations may experience greater flood hazard due to increased water levels. Thus, although floodplain impacts may not be as widespread in the downstream areas of Cypress Creek, new development could exacerbate risk for structures that are already vulnerable to flooding. 4.2 Effectiveness of Site-scale Detention Features In addition to understanding overall watershed sensitivity to development, and identifying vulnerable locations, this study also evaluated the effectiveness of existing detention requirements to mitigate impacts from future development. Based on floodplain extent results, it is clear that on-site detention policies are unable to completely mitigate the impacts of future development. Previous studies have argued that on-site detention systems that are dispersed across a watershed can better alleviate the impacts from new development compared to large regional detentions systems (Fletcher et al., 2013), because they are better able to replicate pre-development hydrologic response (McCuen and Rawls, 1979; Mogollón et al., 2016). Although these systems can mitigate increase in peak flow (Aulenbach et al., 2017), they are generally most effective for intermediate storms rather than extreme precipitation events (Konrad and Burges, 2001).

Roughness coefficients were increased substantially to replicate the effect of on-site detention standards for this area that were intended to preserve pre-development hydrologic conditions. However, the impact of increased imperviousness overwhelmed the study’s on-site detention proxy resulting in increases in floodplain extents that disproportionately affected specific areas. These results suggest that in addition to limiting peak flow rates with on-site detention other hydrologic characteristics should be considered such as limiting runoff volume with on-site retention. Furthermore, the method presented in this study could be used to determine the necessary level of sitescale mitigation necessary to achieve no adverse impacts at a regional scale. Development policies should be crafted by examining multiple spatial scales in order to properly quantify the cumulative impacts of development, and set mitigation criteria that produces no adverse impacts across the entire watershed. The application of an integrated approach for land use projection modeling and hydrologic/hydraulic modeling to the Cypress Creek watershed highlights its usefulness for quantifying the impacts of future development trends and local development policies. By taking into account both where development occurs regionally (through the land use change drivers within the urban growth model) and how development is managed at the site-scale (through representation of site-scale detention within the hydrologic model), this methodology is able to more accurately delineate areas of future risk since it considers both the future drivers of flood hazard and the specific development policies that seek to mitigate this hazard. Consideration of the interaction of between both these factors has not been implemented within a hydrologic modeling framework before, and represents a significant step forward in the evaluation of the long-term impacts of land

development. Floodplain managers and engineers could utilize a similar approach to evaluate the long-term effectiveness of flood infrastructure or different development policies under nonstationary land use. Specifically, by considering the regional effectiveness of existing site-scale development policies, decision-makers at the city/county level can understand whether these policies will be sufficient to mitigate the long-term impacts from incremental new development. More broadly, this approach could be utilized to determine when FEMA floodplain maps should be updated due to significant changes in the floodplain extent, and this can help mitigate the persistence of outdated flood maps. Although many floodplain managers recognize the need to update FEMA maps periodically, in practice it is difficult to determine when a hydrologic region has changed significantly enough to warrant a map revision. This paper advances a method that quantifies the expected growth in the floodplain over several decades, and can consequently provide specific information about when and how often map revisions are needed. 5 Conclusions This paper presented an approach for quantifying the contributions of future urbanization on evolving flood risk by considering the impact of both regional development patterns and site-scale development policies on 100-year floodplain evolution. Specifically, future development scenarios generated from a land use projection model and existing development policies regulated at the county level were represented within a distributed hydrologic model and unsteady hydraulic model to quantify impacts on the 100-year floodplain. Results demonstrated that future development in 2050 will increase the 100year floodplain extent by 8.4%-12.5% across the watershed, and increase the number of

parcels inside the 100-year floodplain by 12.3%-18.8%. Additionally, results from this study indicate that current on-site detention policies, which only regulate peak flow discharges from new developments, are insufficient to entirely mitigate impacts from new development. The results of this study represent an important contribution toward better understanding how future urbanization can impact riverine flood hazard in the Houston region. By developing and implementing a method to represent site-scale detention features within a watershed-scale hydrologic model, this study also provides a valuable methodological contribution illustrating how the effectiveness of local development policies can be evaluated within the context of long-term land use changes. There are still several limitations associated with the modeling approach presented in this paper that should be addressed with future research efforts. One of the primary limitations of the current approach is the method for representing site-scale policies, since it is assumed that all new developments are the same size. In reality, runoff impacts could be considerably different between different sized developments, and the configuration of development (traditional designs vs open space configurations) could also impact runoff response. Another major limitation of the results presented in this study is the uncertainty associated with future land use scenarios. Although the land use projection model was validated against 2011 land use data, the extrapolation of this model to 2050 introduces considerable uncertainty in the future development estimates. The authors have attempted to take this into account by considering a low and high estimate, and thereby capturing a range of potential development conditions, but there is a need for more thorough uncertainty quantification of these scenarios. Future work will focus on addressing these limitations by investigating how different development sizes and configurations could

produce differing runoff response, and by investigating uncertainty quantification methods for the land use projections and hydrologic/hydraulic models to ultimate propagate uncertainty into floodplain impact results. . Despite the limitations mentioned above, the methods and results presented in this paper improve understanding about the extent to which incremental land use changes can significantly alter the reality of flood risk in urbanizing watersheds. This approach could be applied to a range of other riverine systems across the world, and can aid in investigating and evaluating development regulations that could be more effective in mitigating flooding. Acknowledgements This work was supported by the NSF PIRE Grant no. OISE-1545837, and by the Texas Water Resources Institute and U.S. Geological Survey Grant no. 06-M1701864. References Arsanjani, J.J., Helbich, M., Kainz, W., Boloorani, A.D., 2012. Integration of logistic regression, Markov chain and cellular automata models to simulate urban expansion. Int. J. Appl. Earth Obs. Geoinf. 21, 265–275. https://doi.org/10.1016/j.jag.2011.12.014 Aulenbach, B.T., Landers, M.N., Musser, J.W., Painter, J.A., 2017. Effects of Impervious Area and BMP Implementation and Design on Storm Runoff and Water Quality in Eight Small Watersheds. J. Am. Water Resour. Assoc. 53, 382–399. https://doi.org/10.1111/1752-1688.12501 Bass, B., Juan, A., Gori, A., Fang, Z., Bedient, P., 2017. 2015 Memorial Day Flood Impacts for Changing Watershed Conditions in Houston. Nat. Hazards Rev. 18, 1–

11. https://doi.org/10.1061/(ASCE)NH.1527-6996.0000241. Birkland, T.A., Burby, R.J., Conrad, D., Cortner, H., Michener, W.K., 2003. River Ecology and Flood Hazard Mitigation. Nat. Hazards Rev. 4, 46–54. https://doi.org/10.1061/(ASCE)1527-6988(2003)4:1(46) Blessing, R., Sebastian, A., Brody, S.D., 2017. Flood Risk Delineation in the United States : How Much Loss Are We Capturing ? Nat. Hazards Rev. 18, 1–10. https://doi.org/10.1061/(ASCE)NH.1527-6996.0000242. Butt, M.J., Umar, M., Qamar, R., 2013. Landslide dam and subsequent dam-break flood estimation using HEC-RAS model in Northern Pakistan. Nat. Hazards 65, 241–254. https://doi.org/10.1007/s11069-012-0361-8 Castellarin, A., Di Baldassarre, G., Brath, A., 2011. Floodplain Management Strategies for Flood Attenuation in the River Po. River Res. Appl. 27, 1037–1047. https://doi.org/10.1002/rra Choi, W., Deal, B.M., 2008. Assessing hydrological impact of potential land use change through hydrological and land use change modeling for the Kishwaukee River basin (USA). J. Environ. Manage. 88, 1119–1130. https://doi.org/10.1016/j.jenvman.2007.06.001 Doubleday, G., Sebastian, A., Luttenschlager, T., Bedient, P.B., 2013. Modeling Hydrologic Benefits of Low Impact Development: A Distributed Hydrologic Model of The Woodlands, Texas. J. Am. Water Resour. Assoc. 49, 1444–1455. https://doi.org/10.1111/jawr.12095 Du, J., Qian, L., Rui, H., Zuo, T., Zheng, D., Xu, Y., Xu, C.Y., 2012. Assessing the effects of urbanization on annual runoff and flood events using an integrated

hydrological modeling system for Qinhuai River basin, China. J. Hydrol. 464–465, 127–139. https://doi.org/10.1016/j.jhydrol.2012.06.057 Eastman, J.R., 2012. IDRISI Selva Manual. Clark University, Worcester, MA. Eastman, J.R., 2006. IDRISI 15: the Andes edition. Clark University, Worcester, MA. Fang, Z., Zimmer, A., Bedient, P.B., Robinson, H., Christian, J., Vieux, B.E., 2010. Using a Distributed Hydrologic Model to Evaluate the Location of Urban Development and Flood Control Storage. J. Water Resour. Plan. Manag. 136, 597– 601. https://doi.org/10.1061/(ASCE)WR.1943-5452.0000066 Fletcher, T.D., Andrieu, H., Hamel, P., 2013. Understanding, management and modelling of urban hydrology and its consequences for receiving waters: A state of the art. Adv. Water Resour. 51, 261–279. https://doi.org/10.1016/j.advwatres.2012.09.001 Habete, D., Ferreira, C.M., 2017. Potential Impacts of Sea-Level Rise and Land-Use Change on Special Flood Hazard Areas and Associated Risks. Nat. Hazards Rev. 18, 04017017. https://doi.org/10.1061/(ASCE)NH.1527-6996.0000262 Hallegatte, S., Green, C., Nicholls, R.J., Corfee-Morlot, J., 2013. Future flood losses in major coastal cities. Nat. Clim. Chang. 3, 802. HCFCD, 2017. Cypress Creek [WWW Document]. Harris Cty. Flood Control Dist. URL https://www.hcfcd.org/projects-studies/cypress-creek/ HEC, 2016. HEC-RAS River Analysis System User ’ s Manual, USACE. Hollis, G.E., 1975. The effect of urbanization on floods of different recurrence interval. Water Resour. Res. 11, 431–435. https://doi.org/10.1029/WR011i003p00431 Homer, C., Dewitz, J., Yang, L., Jin, S., Danielson, P., Coulston, J., Herold, N., Wickham, J., Megown, K., 2015. Completion of the 2011 National Land Cover

Database for the Conterminous United States – Representing a Decade of Land Cover Change Information. Program. Eng. Remote Sens. 345–354. Hu, Z., Lo, C.P., 2007. Modeling urban growth in Atlanta using logistic regression. Comput. Environ. Urban Syst. 31, 667–688. https://doi.org/10.1016/j.compenvurbsys.2006.11.001 Hupp, C.R., 2000. Hydrology, geomorphology and vegetation of costal plain rivers in the south-eastern USA. Hydrol. Process. 14, 2991–3010. https://doi.org/10.1002/10991085(200011/12)14:16/17<2991::AID-HYP131>3.0.CO;2-H Ji, C.Y., 2000. Land-Use Classification of Remotely Sensed Data Using Kohnonen SelfOrganizing feature Map Neural Networks. Photogramm. Eng. Remote Sens. 66, 1451–1460. Juan, A., Hughes, C., Fang, Z., Bedient, P., 2017. Hydrologic Performance of WatershedScale Low-Impact Development in a High-Intensity Rainfall Region. J. Irrig. Drain. Eng. 143, 1–11. https://doi.org/10.1061/(ASCE)IR.1943-4774.0001141. Kalyanapu, A.J., Burian, S.J., McPherson, T.N., 2009. Effect of land use-based surface roughness on hydrologic model output. J. Spat. Hydrol. 9, 51–71. Kim, H.W., Li, M.H., Kim, J.H., Jaber, F., 2016. Examining the Impact of Suburbanization on Surface Runoff using the SWAT. Int. J. Environ. Res. 10, 379– 390. Knebl, M.R., Yang, Z.L., Hutchison, K., Maidment, D.R., 2005. Regional scale flood modeling using NEXRAD rainfall, GIS, and HEC-HMS/ RAS: A case study for the San Antonio River Basin Summer 2002 storm event. J. Environ. Manage. 75, 325– 336. https://doi.org/10.1016/j.jenvman.2004.11.024

Konrad, C.P., Burges, S.J., 2001. Hydrologic Mitidation Using On-Site Residential Storm-Water Detention. J. Water Resour. Plan. Manag. 127, 99–107. Leopold, L., 1968. Hydrology for Urban Land Planning - A Guidebook on the Hydrologic Effects of Urban Land Use. Geol. Surv. Circ. 554, 1–21. Lin, Y.P., Chu, H.J., Wu, C.F., Verburg, P.H., 2011. Predictive ability of logistic regression, auto-logistic regression and neural network models in empirical land-use change modeling-a case study. Int. J. Geogr. Inf. Sci. 25, 11–23. Lindner, J., Fitzgerald, S., 2016. RE: Immediate Report – Final April 17-18, 2016 (Tax Day) Storm and Flood Information. McCuen, R.H., Rawls, W.J., 1979. Classification of Evaluation of Flood Flow Frequency Estimation Techniques. JAWRA J. Am. Water Resour. Assoc. 15, 88–93. https://doi.org/10.1111/j.1752-1688.1979.tb00291.x McLin, S.G., Springer, E.P., Lane, L.J., 2001. Predicting floodplain boundary changes following the Cerro Grande wildfire. Hydrol. Process. 15, 2967–2980. https://doi.org/10.1002/hyp.385 Michel-Kerjan, E., Lemoyne de Forges, S., Kunreuther, H., 2012. Policy Tenure Under the U.S. National Flood Insurance Program (NFIP). Risk Anal. 32, 644–658. https://doi.org/10.1111/j.1539-6924.2011.01671.x Mogollón, B., Frimpong, E.A., Hoegh, A.B., Angermeier, P.L., 2016. Recent Changes in Stream Flashiness and Flooding, and Effects of Flood Management in North Carolina and Virginia. J. Am. Water Resour. Assoc. 52, 561–577. https://doi.org/10.1111/1752-1688.12408 Montz, B., Gruntfest, E.C., 1986. Changes in American urban floodplain occupancy since

1958: the experiences of nine cities. Appl. Geogr. 6, 325–338. https://doi.org/10.1016/0143-6228(86)90034-2 NLCD, 2011. National Land Cover Database 2011 [WWW Document]. NRC, 2014. Reducing Coastal Risk on the East and Gulf Coasts. National Academies Press, Washington, D. C. https://doi.org/10.17226/18811 Ntelekos, A.A., Oppenheimer, M., Smith, J.A., Miller, A.J., 2010. Urbanization, climate change and flood policy in the United States. Clim. Change 103, 597–616. https://doi.org/10.1007/s10584-009-9789-6 Peterson, A.T., Papeş, M., Soberón, J., 2008. Rethinking receiver operating characteristic analysis applications in ecological niche modeling. Ecol. Modell. 213, 63–72. https://doi.org/10.1016/J.ECOLMODEL.2007.11.008 Pijanowski, B.C., Brown, D.G., Shellito, B.A., Manik, G.A., 2002. Using neural networks and GIS to forecast land use changes: a Land Transformation Model. Comput. Environ. Urban Syst. 26, 553–575. https://doi.org/10.1016/S01989715(01)00015-1 Pinter, N., 2005. One Step Forward, Two Steps Back on U.S. Floodplains. Science (80-. ). 308, 207–209. Pumo, D., Arnone, E., Francipane, A., Caracciolo, D., Noto, L. V., 2017. Potential implications of climate change and urbanization on watershed hydrology. J. Hydrol. 554, 80–99. https://doi.org/10.1016/j.jhydrol.2017.09.002 Qian, Z., 2010. Without zoning: Urban development and land use controls in Houston. Cities 27, 31–41. https://doi.org/https://doi.org/10.1016/j.cities.2009.11.006 Rawls, W.J., Brakensiek, D.L., Miller, N., 1983. Green-ampt Infiltration Parameters from

Soils Data. J. Hydraul. Eng. 109, 62–70. https://doi.org/10.1061/(ASCE)07339429(1983)109:1(62) Ray, T., Stepinski, E., Sebastian, A., Bedient, P.B., 2011. Dynamic Modeling of Storm Surge and Inland Flooding in a Texas Coastal Floodplain. J. Hydraul. Eng. 137, 1103–1110. https://doi.org/10.1061/(ASCE)HY.1943-7900.0000398 Rosburg, T.T., Nelson, P.A., Bledsoe, B.P., 2017. Effects of Urbanization on Flow Duration and Stream Flashiness: A Case Study of Puget Sound Streams, Western Washington, USA. J. Am. Water Resour. Assoc. 53, 493–507. https://doi.org/10.1111/1752-1688.12511 Rose, S., Peters, N.E., 2001. Effects of urbanization on streamflow in the Atlanta area (Georgia, USA): A comparative hydrological approach. Hydrol. Process. 15, 1441– 1457. https://doi.org/10.1002/hyp.218 Rothwell, J., 2012. Housing costs, zoning, and access to high-scoring schools. Sangermano, F., Eastman, J.R., Zhu, H., 2010. Similarity Weighted Instance-based Learning for the Generation of Transition Potentials in Land Use Change Modeling. Trans. GIS 14, 569–580. Sebastian, A., Lendering, K., Kothuis, B., Brand, N., Jonkman, B., 2017. Hurricane Harvey Report: A fact-finding effort in the direct aftermath of Hurricane Harvey in the Greater Houston Region. Delft Delft Univ. Publ. 102. Shafizadeh Moghadam, H., Helbich, M., 2013. Spatiotemporal urbanization processes in the megacity of Mumbai, India: A Markov chains-cellular automata urban growth model. Appl. Geogr. 40, 140–149. https://doi.org/10.1016/j.apgeog.2013.01.009 Sheng, J., Wilson, J.P., 2009. Watershed urbanization and changing flood behavior across

the Los Angeles metropolitan region. Nat. Hazards 48, 41–57. https://doi.org/10.1007/s11069-008-9241-7 Shooshtari, S.J., Gholamalifard, M., 2015. Scenario-based land cover change modeling and its implications for landscape pattern analysis in Neka Watershed, Iran. Remote Sens. Appl. Soc. Environ. 1, 1–19. Storey, A.L., Talbott, M.D., Fitzgerald, S., 2010. Policy, Criteria, and Procedure Manual for Approval and Acceptance of Infrastructure. Suriya, S., Mudgal, B. V., 2012. Impact of urbanization on flooding: The Thirusoolam sub watershed - A case study. J. Hydrol. 412–413, 210–219. https://doi.org/10.1016/j.jhydrol.2011.05.008 Thapa, R.B., Murayama, Y., 2012. Scenario based urban growth allocation in Kathmandu Valley, Nepal. Landsc. Urban Plan. 105, 140–148. https://doi.org/10.1016/j.landurbplan.2011.12.007 Vieux, B.E., Bedient, P.B., 2004. Assessing urban hydrologic prediction accuracy through event reconstruction. J. Hydrol. 299, 217–236. https://doi.org/10.1016/j.jhydrol.2004.08.005 Vogel, R.M., Yaindl, C., Walter, M., 2011. Nonstationarity: Flood magnification and recurrence reduction factors in the united states. J. Am. Water Resour. Assoc. 47, 464–474. https://doi.org/10.1111/j.1752-1688.2011.00541.x Wheater, H., Evans, E., 2009. Land use , water management and future flood risk. Land use policy 26S, 251–264. https://doi.org/10.1016/j.landusepol.2009.08.019 Wijesekara, G.N., Gupta, A., Valeo, C., Hasbani, J.G., Qiao, Y., Delaney, P., Marceau, D.J., 2012. Assessing the impact of future land-use changes on hydrological

processes in the Elbow River watershed in southern Alberta, Canada. J. Hydrol. 412–413, 220–232. https://doi.org/10.1016/j.jhydrol.2011.04.018 Woltemade, J., Potter, W., 1994. A watershed modeling analysis of fluvial geomorphologic influences on flood peak attenuation. Water Resour. Res. 30, 1933– 1942. Zheng, Z., 2011. Growth surges in area in the Cy-Fair area. Houst. Chron.

Streams Flood Control Reservoir Watershed Boundary

2011 Land Cover

Approximate over-flow area

Cypress Creek Watershed

Open Water Natural Agriculture Developed

Addicks Reservoir Watershed

^Houston

^

0

5

10

20 Kilometers

±

Figure 1: Cypress Creek watershed study area and regional land use. During large rain events Cypress Creek overflows its banks at the “Approximate overflow area” location and spills into the Addicks Reservoir watershed, eventually flowing through the three main streams in the watershed and into its flood control reservoir.

5 4 1

2

(b) Calibration Events R2 = 0.96

3

(c) Validation Events R2 = 0.94

Figure 3: (a) Location of flow calibration/validation points and hydrograph comparisons for a sample location in the middle of the watershed, (b) modeled vs observed peak flow for both calibration events (April 2016 and May 2016), and (c) modeled vs observed peak flow for both validation events (March 2016 and August 2017)

List of Figures Figure 1: Cypress Creek watershed study area and regional land use. During large rain events Cypress Creek overflows its banks at the “Approximate overflow area” location and spills into the Addicks Reservoir watershed, eventually

May 2016

! !

Downstream BC Gauge

Absolute Difference (m) 0.00 - 0.30 0.30 - 0.50 0.50 - 0.60 Overflow Area

April 2016

!

August 2017

!

0

5

10

20 Kilometers

±

Figure A1: Absolute difference (in meters) between observed and modeled peak depth at each comparison location for the May 2016 event (top), April 2016 event (middle), and August 2017 event (bottom)

B A

2011 Conditions 2050 Low Development 2050 High Development

A

B

Figure 7: 2011 and 2050 projected 100-year floodplain comparisons, with a zoom in on the overflow area [part A] and a zoom in on a section in the middle portion of the watershed [part B]

2011 Conditions

Open Water Natural Agriculture Developed - Low Intensity Developed - Medium Intensity

2050 Low Development

2050 High Development

0

3.75 7.5

15 Kilometers

±

Figure 6: Land use evolution in Cypress Creek watershed for current conditions (top), 2050 low development scenario (middle), and 2050 high development scenario (bottom). Low intensity development refers to single family residential land uses with significant pervious cover, while medium/high intensity development refers to either concentrated residential development or commercial/industrial development

flowing through the three main streams in the watershed and into its flood control reservoir. Figure 2: Modeling approach overview depicting inputs to each model and linkages between models Figure 3: (a) Location of flow calibration/validation points and hydrograph comparisons for a sample location in the middle of the watershed, (b) modeled vs observed peak flow for both calibration events (April 2016 and May 2016), and (c) modeled vs observed peak flow for both validation events (March 2016 and August 2017) Figure 4: Process flowchart for determining representative roughness value (r0) to model on-site detention features Figure 5: Figure 5: (a) Peak depth validation locations, (b) modeled vs observed peak depth for May 2016 event, (c) modeled vs observed peak depth for April 2016 event, and (d) modeled vs observed peak depth for August 2017 event Figure 6: Land use evolution in Cypress Creek watershed for current conditions (top), 2050 low development scenario (middle), and 2050 high development scenario (bottom). Low intensity development refers to single family residential land uses with significant pervious cover, while medium/high intensity development refers to either concentrated residential development or commercial/industrial development Figure 7: 2011 and 2050 projected 100-year floodplain comparisons, with a zoom in on the overflow area [part A] and a zoom in on a section in the middle portion of the watershed [part B] Figure 8: Comparison of channel morphology between a typical cross section in the (a) upstream overflow area, and (b) midstream portion of the watershed. The increase in 100year water level (in meters) between the two areas is shown, with the range of values indicating the low – high 2050 scenarios *Evidence likelihood transformation Category Driver Table 1: A priori drivers of urbanization Environmental Distance to coast Distance to streams Distance to parks Land cover type* Infrastructure Distance to developed Distance to road network Distance to downtown Distance to schools Socioeconomic Property value Percent employed School district Census Place*

Agriculture Table 2: Percent changes in land use type between 2011 38.7% 17.6% 43.7% 2050 projections and current conditions 2050 (low) 53.0% 8.0% 38.9% Table 3: Floodplain extent increase and 2050 (high) 59.8% 6.8% 33.4% inundated parcels Percent change increase for entire 37.0% -54.5% -11.0% from 2011 (low) Cypress Watershed Percent change Table 4: Floodplain 54.5% -61.4% -23.6% from 2011 (high) extent increase 2011 2050 (low) 2050 (high) and inundated parcels increase Floodplain Extent (km2) 106.5 115.5 119.8 for overflow Percent Increase from region of Cypress 8.4% 12.5% 2011 Creek watershed Inundated Parcels 2926 3287 3476 Highlights  Evaluatio Percent Change from 12.3% 18.8% n of future 2011 urbanization 2011 2050 (low) 2050 (high) impacts on Floodplain Extent (km2) 38.5 44.8 47.3 floodplain Percent Increase from considering 16.4% 23.0% 2011 regional and local Inundated Parcels 413 499 515 development Percent Increase from characteristics 20.8% 24.7% 2011  Combined land use projection through machine learning model with hydrologic and hydraulic modeling  Floodplain increases of up to 12.5% under projected urbanization scenarios, and existing development policies are unable to mitigate urbanization impacts Developed

Natural