Sediment routing in an incised valley during Hurricane Harvey (2017) in Houston, Texas, USA: Implications for modern sedimentation

In August of 2017, Hurricane Harvey brought ~6.2 km³ of precipitation to Houston, Texas (USA; Fig. 1; Blake and Zelinsky, 2018), flooding more than 150,000 homes (Linder and Fitzgerald, 2018). More than 11 km³ of runoff entered fluvial-estuarine systems in the Houston area, mobilizing millions of cubic meters of sediment (Du et al., 2019). Although the region has experienced flooding in modern times (during the past 200 yr), no previous study has documented the details of sediment flux through each individual drainage. The largest sediment volumes are mobilized and deposited within drainage banks during bankfull discharges, the point at which the water level fills the channel but does not exceed the banks, reducing future stream flow capacity and raising susceptibility to flooding (Wu et al., 2008). Frequency of bankfull flows have risen in Houston due to increasing coverage of impervious surfaces that raise surface runoff volumes (Khan, 2005). Improving the understanding of sediment routing and associated sedimentation during bankfull flows, such as induced by flooding from a tropical cyclone, allows for contextualization of modern flooding events to ones from the Holocene and earlier (Liu and Fearn, 1993).

Houston drainages are tributaries to Galveston Bay that formed in the Pleistocene from eastward deflection of Brazos River meander-belt ridges, becoming progressively younger to the south (Fig. 1; Dupré, 2019). Galveston Bay is one of several underfilled incised valleys along the Texas coast (Anderson et al., 2016), characterized by a backstepping succession of fluvial and estuarine facies since the Stage 2 (last glacial maximum) sequence boundary ca. 18 ka (Rodriguez et al., 2005). Throughout the Holocene, the San Jacinto, Trinity, and other Galveston Bay tributaries delivered ~2.3 × 10⁵ to 9.0 × 10⁵ m³/yr of sediment, insufficient to fill Galveston Bay (Anderson et al., 2016; Table S1 in the Supplemental Material¹).

In modern times, Houston drainages are fluvial-estuarine systems, with tidal influence extending several miles upstream (Fig. 1). These drainages have been highly modified through channel straightening, removal of bank vegetation, and paving of the banks (USACE, 2020). Fluvial sediment delivery to Galveston Bay has increased 2–5 fold to ~5.3 × 10⁵–4.53 × 10⁶ m³/yr, (Table S1), attributable to higher erosion rates from land development and agricultural practices (Kemp et al., 2020), but remains inadequate to fill Galveston Bay as relative sea level continues to rise (Miller and Shirzaei, 2019).

Significant modifications were made to the Houston drainage system in 1945 and 1948 CE with the construction of flood-controlled Addicks and Barker reservoirs, respectively, ~50 km west of downtown Houston (Fig. 1; USACE, 2009). Since dam construction, developed land area has grown from <1% to over 40% in Addicks and Barker watersheds (Li et al., 2020). Currently, ~20,000 houses occupy the shadow of the reservoir dams (USACE, 2020), ~8000 of which flooded during Hurricane Harvey (Linder and Fitzgerald, 2018).

In the geologic record, fluvial sediment delivery to a basin is often characterized as pulsed rather than continuous. Blum and Törnqvist (2000) abstracted this concept in incised valleys with two end members: (1) a “vacuum cleaner” model (VCM) with sequestered sediment delivered in a pulse to the basin during extreme events, such as a major flood; and (2) a “conveyor belt” model (CBM) where sediment is continuously delivered to the basin. A VCM is more applicable to an underfilled incised valley, while a CBM correlates more often to an overfilled incised valley.

We determined the net minimum volume of sediment transported through a highly modified urban drainage system in Houston, Texas (Fig. 1), during Hurricane Harvey, and tested the pertinence of a VCM. We carried out an integrated quantitative analysis using subaerial elevation change, ground-based observations, satellite and historical aerial imagery, and published studies. Regionally extensive digital elevation models (DEMs), made available from recently acquired lidar data, have made a comprehensive study possible for the first time.

Sediment transport at a regional scale was quantified using subaerial elevation change, before and after Harvey, along drainages via infrared single-wavelength derived DEMs resolved to ≤1 m. Elevation change data sets were generated for each drainage to visualize deposition and erosion, and artificial values from anthropogenic structures or flow level disparities were removed using historical imagery (Fig. S7).

Uncertainty was quantified at three scales from largest to smallest: (1) a regional scale by direct comparison of areas with high and low expected elevation change, (2) an instrument scale via a signal-to-noise sensitivity analysis, and (3) a stream scale using a consistent uncertainty at the scale of interpretation (Figs. S14–S16, Table S7 [see ^{footnote 1}]).

To investigate sediment transport with respect to drainage gradient, elevation profiles were generated for the 12 stream drainages and feeder creeks of Addicks and Barker reservoirs (Figs. 3 and 4). Additionally, given the presence of artificial channel modifications within the drainage network, we classified four types of streams (natural, straightened natural, straightened grass, and straightened concrete) using historical imagery and sinuosity (Fig. S12).

Net minimum sediment transport volume, V_T, totaled ~2.723 × 10⁷ m³ using Equation 2, with the stream drainages recording ~7.98 × 10⁶ m³ (Fig. 2A). In upstream locations, post-Harvey ground images showed exposed tree roots from bank erosion and low in-channel and overbank deposition, consistent with areas of elevation gain and loss (Fig. S9A). Downstream, post-storm ground images showed meter-scale sand bars, in line with a higher degree of in-channel and overbank deposition exhibited by positive elevation change (Fig. S9B). This equates to an average of ~61,000 m³/km² of sediment above the tidal limit, occurring on average within 0.33 km of the lowest observable break in slope, compared to ~78,000 m³/km² below (Fig. 3A). Therefore, sediment was sourced from higher elevations west-northwest of Houston to lower elevations toward Galveston Bay.

Addicks and Barker reservoirs totaled ~7.15 × 10⁶ m³ of net minimum sediment transported (Fig. 2A). At the upper limit of the reservoirs, crevasse splays and moderate overbank deposition were observed along the feeder streams (Fig. 3B). The mid-upper and mid-lower portions of the reservoir resembled sediment transport in the stream drainages with upstream erosion and downstream deposition, while channel scour characterized the feeder streams in the lowermost portions (Fig. 3B). This indicates that large volumes of sediment were initially deposited at the reservoir limit and then reworked in two stages as the reservoirs were drained over a span of 45 days following Harvey (Fig. 3B).

The four stream types we identified in this study (natural, straightened natural, straightened grass, and straightened concrete; Fig. 4-15) showed progressively increasing sediment bypass as the degree of artificial channel modification increased (Fig. 4-15; Table S6). Natural streams exhibited higher in-channel and overbank deposition, while straightened natural, straightened grass, and straightened concrete streams demonstrated a continuing rise in bank erosion and decrease in-channel and overbank deposition (Figs. 4-1–4-14).

The 4314 km² across the 14 drainages recorded ~6.2 km³ of precipitation from Harvey (Fig. 1), transporting a minimum of ~2.723 × 10⁷ m³ of sediment, V_T (Fig. 2). This volume is ~6–51 times larger than the average annual volume of sediment delivered to Galveston Bay in modern times, and ~30–118 times larger when compared to Holocene rates (Table S1). For comparison, V_T is equivalent to ~40% of the average annual sediment volume delivered by the Mississippi River to the Gulf of Mexico (Fig. 1; Meade and Moody, 2010).

Dammed reservoirs are source-to-sink systems and a barrier to sediment transport, causing reservoir capacities to decrease over time (Kondolf et al., 2014). Addicks and Barker capacities were reduced by ~1.2% and ~1.6%, respectively, from pre-Harvey capacities (USACE, 2009; Fig. 2A) from sediment deposition near the reservoir limit and mid-lower portions coupled with a widespread floodplain deposit (Fig. 3B). As urban sprawl raises the runoff ratio from the increasing impervious area (Li et al., 2020), progressively higher sediment volumes will be transported and deposited in the reservoirs, decreasing capacities. Thus, flood risk is projected to rise for ~20,000 homes in the shadow of the two dams (USACE, 2020).

As Hurricane Harvey approached the Houston area for a second landfall (Fig. 1), depressed windspeeds muted the storm surge (Blake and Zelinsky, 2018), reducing associated landward sediment transport. Instead, initial rainfall on 25 August induced seaward sediment transport shown by ~0.5 m and ~1.3 m scours in lower portions of the San Jacinto River and Buffalo Bayou, respectively (Du et al., 2019; Dellapenna et al., 2020), ~0.2–0.8 m erosion at San Luis Pass, and widespread erosion at Bolivar Roads Inlet (Fig. 1; Ramón-Dueñas et al., 2021). Ongoing rainfall on 26–30 August generated ~11 km³ of runoff (Du et al., 2019), causing a freshwater piling effect in Galveston Bay and lower parts of its tributaries (Huang et al., 2021). Ebb flows reached 3 m/s in the ~3-km-wide Bolivar Roads Inlet, a condition that produces seaward-directed sediment transport (Goff et al., 2010), reflected by a sediment plume extending ~60 km offshore opposite the normal southwest longshore current (Ramón-Dueñas et al., 2021). Flow velocities subsided as precipitation tapered off on 31 August (Du et al., 2019), predominantly stranding sediment in downstream navigable channels, including the HSC (Dellapenna et al., 2020), and subaerially, along drainage banks below the tidal limit and break in slope (Fig. 3A).

Sediment routing during Hurricane Harvey resembled a VCM, as described by Blum and Törnqvist (2000). Since the Pleistocene, Houston drainages incise during normal flows (Dupré, 2019). However, during high flows, these systems initially bypass alluvial sediment carried by floodwaters. Over the past 200 years across North America, alluvial sediment volumes have increased 10-fold due to land development (Kemp et al., 2020); Houston developed land area grew by 30% from 1997 to 2017 (Hakkenberg et al., 2019). As floodwaters recede, the system shifts to both in-channel and overbank deposition, which increases in magnitude downstream as sediment falls out of suspension (Fig. 3). The resulting geometry decreases the volume of future bankfull flows and creates a sediment-choked system, increasing susceptibility to flooding in the surrounding area. The west fork of the San Jacinto River entering Lake Houston depicts this scenario, where ~1.38 × 10⁶ m³ of a mouth bar was dredged following Harvey to reduce flooding occurrences in the surrounding area (Fig. 1; USACE, 2018).

The region is also facing accelerated relative sea-level rise from 34 mm/yr to 49 mm/yr of subsidence in low-lying areas (Miller and Shirzaei, 2019) coupled with eustatic sea-level rise of 4–20 mm/yr (Magnan et al., 2022). Rising water translates the tidal limit and break in slope landward toward the Stage 5 shoreline, ~1–5 m above present sea level (Figs. 1 and 3A; Simms et al., 2013), relocating areas of high deposition upstream to additional urban areas.

Comparison of sediment volumes transported through natural and modified drainages through Houston demonstrates that channel modification resulted in significant bypass of sediment downstream (Fig. 4). Following Hurricane Harvey, the HSC was covered with 1–5 m of sediment (van Maren et al., 2015; Du et al., 2019), and meter-scale sandbars were deposited in downstream segments of Buffalo and Greens Bayous, including ~18,000 m³ of sediment that was later removed from BBP (Fig. 1; BBP, 2018; Kendall et al., 2019). A total of ~6.80 × 10⁶ m³ of sediment was dredged from Texas navigational channels after Harvey (U.S. Congress, 2018), amounting to ~US$351 million (USACE, 2019). At a cost of ~US$500 million per meter of water depth (Center for Ports and Waterways, 2010), and an ongoing HSC expansion (USACE, 2019), it is vital to quantify the systemwide effects of channel modification.

Our integrated analysis quantified the magnitude and drivers of sediment routing during Hurricane Harvey in Houston, Texas, USA. A total of ~6.2 km³ of precipitation mobilized a minimum of ~2.723 × 10⁷ m³ in the Houston area. This volume is ~6–51 times larger than the average annual volume of sediment delivered to Galveston Bay in modern times, and ~30–118 times larger when compared to Holocene rates. Addicks and Barker reservoir capacity decreased by ~1.2% and ~1.6%, respectively. The risk of future flooding increases as reservoir capacity is decreased due to sediment deposition. Sediment was mobilized from west-northwest of Houston and pulsed toward Galveston Bay during Harvey, resembling a VCM. Pulsed sediment delivery creates a sediment choked system following an extreme event, increasing the susceptibility of surrounding areas to future flooding. As more urban areas are subjected to flooding due to the interplay between eustatic sea-level rise, increased rates of precipitation, and the spread of urban development into low-lying areas, details of efficient ways to move both water and sediment through different types of drainage networks is necessary. Natural channels impact urban areas during floods by retaining sediments, while artificial channels bypass sediment downstream to navigable waterways. Dredging is capital-intensive, and structures continue to be built on floodplains, emphasizing the need to quantify the systemwide effects of channel straightening.

We thank William Dupré, Will Sager, Carolina Ramón-Dueñas, and Jon Pelletier for their background insights on the region and their assistance with the project. We thank editor K. Benison, R. Mahon, and two anonymous reviewers whose comments improved this manuscript.