Measuring ephemeral gully erosion rates and topographical thresholds in an urban watershed using unmanned aerial systems and structure from motion photogrammetric techniques

Abstract

Both rural and urban development can lead to accelerated gully erosion. Quantifying gully erosion is challenging in environments where gullies are rapidly repaired, and in urban areas where microtopographic complexity complicates the delineation of contributing areas. This study used unmanned aerial vehicles (UAVs) and Structure-from-Motion (SfM) photogrammetric techniques to quantify gully erosion in the Los Laureles Canyon watershed, a rapidly urbanizing watershed in Tijuana, Mexico. Following a storm event, the gully network extent was mapped using an orthomosaic (0.038 m pixel size); the local slope and watershed area contributing to each gully head were mapped with a Digital Surface Model (0.3 m pixel size). Gullies formed almost exclusively on unpaved roads which had erodible soils and concentrated flow. Management practices (e.g. road maintenance that fill gullies after large storms) contributed to total sediment production at the watershed scale. Sediment production from gully erosion was higher and threshold values of slope and drainage area for gully incision were lower than ephemeral gullies reported for agricultural settings. This indicates high vulnerability of unpaved roads to gully erosion which is consistent with high soil erodibility and low critical shear stress measured in the laboratory with a mini jet-erosion-test device. Future studies that evaluate effects of different soil types on gully erosion rates for unpaved roads, as well as those that model effects of management practices such as road paving and their impact on runoff, soil erosion, and sediment loads are needed to advance sediment management and planning in urban watersheds.

Introduction

In New Zealand, serious problems exist in the North Island’s East Coast region, where the widespread deforestation of soft-rock hill country by European settlers over the last century (Allsop, 1973) has led to severe hillslope erosion and soil loss, and consequent flooding and sedimentation downstream (Trustrum et al., 1999; Page et al., 2001).

Gully erosion is often associated with land degradation caused by anthropogenic activities and is commonly related to changes in catchment hydrology, such as removal of native vegetation and soil disturbance (Oygarden, 2003). Gully characteristics and erosion rates have been well documented in agricultural environments, but high erosion rates are also observed in urban areas during construction (Wolman, 1967). Erosion rates usually decline as the urban landscape matures (Archibold, Levesque, de Boer, Aitken, & Delanoy, 2003), but this is not always the case in developing countries where soil exposure can last for decades following urbanization (Biggs, Atkinson, Powell, & Ojeda, 2010). This can result in chronically high gully erosion rates compared with undisturbed areas or urban areas with high impervious and/or vegetation cover.

There are few studies assessing gully erosion in urban settings (Castillo & Gómez, 2016). Adediji, Jeje, and Ibitoye (2013) described the relationship between urban land surface characteristics and gully erosion in Nigeria and found a significant relationship between soil texture and land use. Other studies describe the expansion and headcut retreat of permanent urban gullies. Archibold et al. (2003) surveyed two urban gullies from 1994 to 2000 and measured gully headcut retreat, widening, and deepening. They attributed gully erosion to land-use change, as was also found by Guerra and Hoffman (2006) in Brazil and by Imwangana, Dewitte, Ntombi, and Moeyersons (2014) in Congo. Nevertheless, little has been done to characterize erosion rates for a large network of gullies, due in part to difficulty in surveying them.

Ephemeral gullies can be important contributors to sediment production at the watershed scale (Vandaele, Poesen, Govers, & van Wesemael, 1996), especially in arid and semiarid environments. Such gullies are small eroded channels formed by concentrated runoff during a storm event (Foster, 1986) and are temporary features removed by tillage operations (Poesen & Govers, 1990) or filled with sediment in urban environments. Ephemeral gullies form from a complex interaction between physical and management attributes such as topography, rainfall duration and intensity, soil moisture, soil properties, vegetation cover, and management practices (Momm, Bingner, Wells, & Wilcox, 2012). An ephemeral gully will form only when a particular interaction between these attributes, or an abrupt change in ground surface elevation, leads to shear stress caused by overland runoff exceeding the soil critical shear stress to produce scour below the soil surface and eventually incision or upstream migration of the gully head (GH). A GH is the location at which gully erosion processes cannot continue upstream under particular combination of rainstorm intensity, storm discharge, land use, vegetation cover, and soil type (Torri & Poesen, 2014).

Several studies have documented the relationship between ephemeral gully formation and runoff erosivity using topographic attributes, especially drainage area (Ad) and local slope (S) at the GH, with a threshold defined by S = aA_d^b (Foster, 1986; Montgomery & Dietrich, 1994; Patton & Schumm, 1975; Thorne, Zevenbergen, Grissinger, & Murphey, 1986). In this equation, ‘a’ is a constant that varies with lithology, soils, climate, and vegetation cover (Vandaele et al., 1996), and “b” is an exponent related to the dominant processes that form the gully (Gómez-Gutiérrez, Schnabel, & Lavado-Contador, 2009). Montgomery and Dietrich (1994) suggested that b = 0.5 for laminar flow and b = 0.86 for turbulent flow. S has been measured using different methods such as clinometer, compass, topographic maps, and digital elevation models (DEMs). Recently, three-dimensional photo-reconstruction techniques have been used to estimate slopes in soil erosion studies (Castillo, James, Redel-Macías, Pérez, & Gómez, 2015; Di Stefano, Ferro, Palmeri, & Pampalone, 2017; Gómez-Gutiérrez, Schnabel, Berenguer-Sempere, Lavado-Contador, & Rubio-Delgado, 2014; Nadal-Romero, Revuelto, Errea, & López-Moreno, 2015) but have not been applied in characterizing ephemeral gullies in an urban context.

Ephemeral gullies can be identified and characterized using aerial imagery because they commonly have distinct color, texture, and shadow characteristics (Gómez-Gutiérrez et al., 2009) that differ from the rest of the landscape. Aerial imagery complements ground surveys of gully erosion, especially for rapidly urbanizing watersheds in developing countries, where ground surveys are difficult due to formation of large gullies on unpaved roads, stream channel bank collapse, landslides, and flooding that impede field access. Urban gullies are often filled in within days of formation and require rapid assessment after a storm. Remote sensing supported by ground control points (GCPs) is often quicker and covers a larger area than manual surveys. GCPs are critical to scale and georeference the remote sensing-derived cartographic products.

Many studies use time series of aerial photographs to map gully erosion, mostly in natural and agricultural areas (Nachtergaele & Poesen, 1999; Parkner, Page, Marutami, & Trustrum, 2006; Ries & Marzolff, 2003). DEMs derived from aerial imagery have also been used (Martínez-Casasnovas, Ramos, & Ribes-Dasi, 2002; Thorne et al., 1986). More recently, structure from motion (SfM) has been used in geomorphic studies, including different spatial scales, environments, and applications (Di Stefano et al., 2017; Gómez-Gutiérrez et al., 2014; James & Robson, 2012; Micheletti, Chandler, & Lane, 2014; Turner, Lucieer, & Watson, 2012; Westoby, Brasington, Glasser, Hambrey, & Reynolds, 2012). SfM is a photogrammetric technique based on computer visualization tools and image-based, three-dimensional surface reconstruction algorithms (James & Robson, 2012). SfM creates massive point clouds based on pixel matching from which highly accurate digital surface models (DSMs), DEMs, and orthophotos can be derived. Accuracy of SfM in assessing gully erosion is very similar to the most accurate topographic methods such as terrestrial laser scanning or traditional photogrammetry (Castillo et al., 2012; Fonstad, Dietrich, Courville, Jensen, & Carbonneau, 2013; James & Robson, 2012), but SfM is cheaper and faster (Castillo et al., 2015; Westoby et al., 2012). High spatial resolution orthophotos derived from SfM can be used to identify the location of gully networks and key characteristics of gully formation such as S and A_d and their relationship with soil loss within a gully network (Gómez-Gutiérrez et al., 2014).

Gully networks, whether in agricultural or in urban areas, have characteristic patterns of sediment production pertaining to A_d. Large watersheds usually have low normalized (per unit area) sediment production compared with small watersheds, because they have more storage capacity to retain sediment and the average slope decreases with increasing watershed size (Castillo & Gómez, 2016; Walling, 1983); this has also been reported for gully erosion rates (Poesen, Nachtergaele, Verstraeten, & Valentin, 2003; Vanwalleghem, Poesen, Nachtergaele, & Verstraeten, 2005). Poesen et al. (2003) and Castillo and Gómez (2016) compiled gully erosion rates for different agricultural environments, but ephemeral gully erosion rates and topographic thresholds for gully initiation in urbanized watersheds are not available.

This paper aims to map gullies and quantify gully erosion rates using unmanned aerial systems (UASs)-based SfM technology to aid understanding of the processes of gully formation in Los Laureles Canyon Watershed (LLCW), a rapidly urbanizing watershed that drains into the Tijuana Estuary in the western section of the United States–Mexico border. It addresses three questions aligned with the objectives: (a) Where do gullies form in an urbanizing landscape in a developing country context? (b) What are the management implications for the control of sediment production? and (c) How do the topographic thresholds for gully formation (S and A_d) and sediment production rates from urban gullies compare with gullies in agricultural settings? The study is novel in presenting topographic thresholds for ephemeral gully formation and erosion rates in an urban environment, using a combination of UASs–SfM photogrammetric techniques.

Methods

Study Area

The San Bernardo (SB) neighborhood (20 ha, Figure 1) is located in the Los Laureles Canyon Watershed (LLCW), a bi-national watershed that flows from Tijuana, Mexico, into the southwestern arm of the Tijuana Estuary, United States. (Figure 1). The climate is Mediterranean, with a wet season from November to April and annual precipitation of ~24 cm/yr. Most storms occur in winter. SB is located on the San Diego formation, which includes deposits of erosive and loosely consolidated sandstone and siltstone, with average slope of 15 degrees. Excessive erosion, transport and deposition of sediment in the LLCW watershed have had many detrimental impacts on the people living in the watershed and have impaired ecosystems in the Tijuana Estuary (Weis et al., 2001).

Figure 1.

a) Geographic location of Los Laureles Canyon Watershed (LLCW) and San Bernardo (SB), b) one example of land degradation caused by gully erosion in Tijuana, Mexico, c) excessive sedimentation in the Tijuana Estuarine Reserve (TJE), USA.

Image acquisition and processing

Both ground- and UAV-based surveys of a gully network that formed following a large storm event on January 5-7, 2016 were conducted on January 16, 2016 (Figure 2). The storm was the largest of the water year (~50 mm), and had a 15-minute maximum intensity of 4.8 mm, which has a 1-year recurrence interval. Other storms occurred during the year, but were smaller than the threshold precipitation typically required to produce gullies in SB (~25-35 mm), as observed on other field visits following storm events during three hydrological years (2013-2016). The volume of sediment generated by the gullies during the January storm was therefore used as the annual total for comparison to other studies. The 2016 water year (October 2015-September 2016) was drier than normal (155 mm total precipitation vs long-term average of 238 mm), so our results likely underestimate the long-term mean sediment production from gullies in this location.

Figure 2.

Daily rainfall time series for the 2016 water year. The gray box represents the rainfall threshold (~25–35 mm) typically needed for gully formation in the study area.

Aerial imagery was acquired during two flights over the SB neighborhood. Digital true color images were acquired using a nonmetric, commercially available digital camera (GoPro, 12 megapixels resolution) mounted on a UAV (DJI Phantom 2). The camera was set to time-lapse capture mode (1 image per second) and images were acquired at an altitude of 75 m (above ground level) to ensure 75% side overlap and 75% forward overlap. The camera mounted facing 15 degrees from vertical to avoid doming deformations (James & Robson, 2014).

SfM was used to create a DSM using 7 GCPs (calibration points, Table S1) and 6 error control points (ECPs or validation points, Table S2) spatially distributed over the study area (Figure 3) and surveyed using differential GPS (Magellan Pro Mart 3) with sub-centimeter to 5 cm vertical accuracy (Magellan Systems Corporation, San Dimas, USA). Other researchers have documented that using 4 to 5 GCPs with additional ECPs could be considered to determine relatively small but widely distributed surface changes (James, Robson, d’Oleire-Oltmanns, & Niethammer, 2017) and that root mean square errors (RMSEs) can be reduced by placing GCPs on the image perimeter (Vericat et al., 2009). A dense point cloud (11 points/m²) was generated using the Agisoft Photoscan Professional software (Agisoft LCC, Russia, Version 1.3.0) from which an orthophoto (0.038-m spatial resolution) and a DSM (0.3-m spatial resolution) were created. The spatial resolution (0.3 m) was calculated as √(1/d), where d is the SfM-derived point density, which resulted in one point, on average, in every 0.3 × 0.3 m pixel.

Figure 3.

Structure from motion-derived orthophoto and location of the study watersheds, ground control points (GCPs), error control points (ECPs), gully heads, and sampling locations

The error of the DSM georegistration was measured as the vertical and horizontal difference between validation (ECP) coordinates and corresponding X–Y or Z coordinate values from the DSM. The RMSE for ECPs was calculated as

$$\mathit{RMSE}=\sqrt{\frac{1}{n}{\Sigma }_{i=1}^n{(\mathrm{dGPScoordinate}\_i-\mathrm{DSMcoordinate}\_i)}^2}$$ (1)

where i is the index of the GCP or ECP points, and ‘n’ is the number of GCPs (7) or ECPs. Additionally, geometry of the orthophoto was tested comparing lengths of 10 different elements (sewer structures) in the image to field data (Table S3). This was deemed as the most appropriate measure of accuracy since the goal was to map dimensions of the gullies, not determine their absolute position.

Topographic thresholds

GHs were identified using the orthophoto (Figure 3), and their topographic attributes (S and A_d) were measured using DSM data. Flow paths and watershed boundaries for each GH, and for various locations along the gully network, were delineated using Hydrology tools in ArcGIS 10.2. A GH was defined as a channel at least 30 cm wide in the uppermost stream cross section. GHs are defined operationally for a given purpose; here, we used 30 cm based on the resolution of the imagery and the width of GH observed in the field. S was calculated as the slope gradient of the DSM-derived flowpath over a distance of 2 m upstream from each GH. A_d was the area draining into each GH calculated from the SfM-derived DSM using the Hydrology tools in ArcMap 10.2. The threshold combination of S and A_d that generates gullies was determined by fitting the equation S = aA_d^b to the lower envelope of the S–A_d plot.

Our method to calculate S follows previous gully erosion studies that use DSMs to calculate upslope lengths (i.e., Gómez-Gutiérrez et al., 2014). Our use of 2 m for the upslope length is similar to Vandekerckhove, Poesen, Oostwould Wijdenes, and de Figueiredo (1998), who used a range from 2 to 4 m. Most studies do not use a constant upslope length, but rather use the nearest two contour lines upslope of each GH to define the upslope segment, and do not report an upslope length (Table S4). In order to determine the sensitivity of a and b to upslope length, both a and b were also determined using upslope distances of 2, 3, and 5 m to calculate S (see Supporting Information).

Specific Soil Loss (SSL)

Fourteen watersheds (Figure 3) were delineated using the watershed tool in ArcGIS 10.2, and were used to estimate gully erosion rates. Watershed outlets were defined by the downslope terminus of the gully network, and watershed sizes spanned the range generally observed in SB. Gully widths and depths were hand-measured in the field, with a ruler, at 48 locations to calibrate and validate the gully dimensions estimated by remote sensing. Polygons of gully perimeter were delineated manually by visual interpretation of the orthophoto. Gullies were too narrow and deep to estimate depth from the DSM, so a uniform depth was estimated for each polygon using visual interpretation informed by the field measurements. The volume eroded was calculated as the product of the polygon area times estimated depth. The specific soil loss (SSL, the average depth of soil loss in the watershed), was then calculated for each watershed as the total gully erosion (m³) normalized by A_d (Figure 4).

Figure 4.

Example of field measurements for watershed 1 (a) and measurement locations, digitized gullies, watershed boundary and outlet used to estimate specific soil loss (b).

Mini-jet test analysis

Five soil samples were collected at representative sites under different land cover conditions (natural, gully walls and filled roads) over the study area to estimate the critical shear stress and soil erodibility using a mini-jet-erosion test following Hanson (1990) and Al-Madhhachi et al. (2013).

Results

SfM-derived DSM

The mean absolute error between observed and modeled object lengths was 2 cm (Table S3), which is 7% of the smallest features we measured, and is the most appropriate error statistic for mapping gully dimensions. The RMSE for ECPs was 3 cm in the X–Y and 7 cm in the Z (Table S2). The SfM-derived DSM had relatively similar (i.e., lower) errors compared with other SfM applications (Dietrich, 2016; Javernick, Brasington, & Caruso, 2014; Vericat et al., 2009). This magnitude of error is acceptable for the smallest features we mapped, which were approximately 30-cm-wide gullies. The relative precision ratio (measurement precision: observation distance) was 1:833, which is similar to 1:950 reported by James and Robson (2012) to evaluate three-dimensional photo reconstruction quality in SfM applications.

Topographic thresholds

A total of 30 GHs were identified by the SfM-derived orthophoto. S correlated inversely with A_d at the GHs (Figure 5), which is consistent with previous studies (Castillo & Gómez, 2016; Kakembo, Xanga, & Rowntree, 2009; Montgomery & Dietrich, 1992; Torri, Sfalanga, & Chisci, 1987; Vandaele et al., 1996; Vanwalleghem et al., 2005). The line fitted through the lower envelope of the S–A_d relationship depicts threshold values required for gully formation (Patton & Schumm, 1975). For instance, an S value of 0.015 and/or an A_d of 0.0008 ha (8 m²) are needed to initiate gully erosion under the rainfall and erodibility conditions in the study area (Figure 5).

Figure 5.

Topographic thresholds (S and A_d) for the mapped headcuts in SB and from selected papers from literature; Colorado (Patton & Schummm, 1975); Oregon and California (Montgomery & Dietrich, 1988); Belgium and Portugal (Vandaele et al., 1995). The line fitted through the lower most points represents the critical conditions for gully initiation.

The mean topographic attributes equation (S = 0.02* A_d ^−0.36) was similar to the one reported by Vandaele et al. (1996) for agricultural settings in central Belgium (S = 0.02* A_d ^−0.40) under a higher precipitation regime. The spatial distribution of the SB point cloud and the values of the a and b coefficients were not sensitive to different upslope lengths (2, 3, and 5 m) defining S: coefficient a was 0.02, 0.0175, and 0.0158 and b was −0.360, −0.355, and −0.387 for upslope lengths of 2, 3, and 5 m (Figure S2), suggesting that both the topographic thresholds for gully formation and the physical characteristics described by these parameters in the system are robust to the upslope length.

Each dataset from the literature that we compared with our data on topographic thresholds (Figure 5) used different cartographic products to measure Ad and S (especially upslope lengths). Vandekerckhove et al. (1998) used an upslope length that range from 2 to 4 m, which is comparable with our upslope length of 2 m, and sensitivity analysis on our data suggests that the topographical thresholds plot and regression coefficients a and b are insensitive to upslope lengths between 2 and 5 m (Figure S2). We conclude that the range of techniques and upslope lengths used by others does not complicate the comparison of results among studies.

Specific soil loss

A total of 311 polygons representing ephemeral gullies within 14 watersheds were identified and mapped from the orthophoto in the SB area. The width measured on the orthophoto ranged from 0.3 to 3.1 m, and depth measured in the field ranged from 0.1 to 0.8 m. SSLdecreased with increasing A_d in SB across two orders of magnitude in A_d (0.04 to 4 ha). SSL in SB was approximately 2–5 times higher than the mean of the sites from Castillo and Gómez (2016), which included sites with precipitation ranging from 40 to 215 mm (Table S4).

Mini-jet test analysis

The critical shear stress values obtained from the jet erosion test ranged from 1.4 to 1.9 Pa (1.6 Pa average), and soil detachment coefficient varied from 324 to 879 cm³/Ns, suggesting high erodibility or low soil resistance values, according to Hanson’s (1990) soil classification diagram. This is consistent with the difference between regression lines reported in this analysis, compared with area-SSL line from Castillo and Gómez (2016), with both indicating high vulnerability to gully erosion.

Discussion

Gully mapping

The study of gullies in rapidly urbanizing regions is particularly challenging due to (a) the ephemeral nature of the gullies, which may be filled in by maintenance crews within hours or days of a storm event, precluding use of historical aerial photographs to map them; and (b) the difficulty in delineating the watershed area draining to a given gully due to the complex drainage networks that form in urban areas. Although existing DEMs can be used to delineate watershed boundaries, they may become outdated as topography changes or have insufficient spatial resolution to accurately identify flow paths and watershed boundaries in a rapidly urbanizing environment. Microtopographic features such as roads, curbs, and ditches can change flow paths and watershed boundaries, and their relative impact on hydrology can be bigger in small watersheds. Both challenges require rapid, high-resolution mapping due to regular management practices implemented on roads that have impacts on gully networks and their contributing areas. UASs–SfM imagery and technology were used to address both challenges in SB.

Gullies formed almost exclusively on roads, reflecting their role in routing flow and their vulnerability to incision. Most gullies in SB are discontinuous because they are usually filled after storms, and the farthest downstream sections have lower slope and are more compacted due to vehicle traffic on the roads, all of which discourage gully formation.

Adediji et al. (2013) reported urban gullies on roads and adjacent to discontinuous concrete channels in Nigeria; gullies in other urban settings were formed in specific sites where runoff is concentrated (Archibold et al., 2003; Imwangana et al., 2014). Roads are the major component of anthropogenic sediment generation in the study area, which has also been reported in other settings, including logged forest (Montgomery & Dietrich, 1988; Reid & Dunne, 1984) and tropical islands (Ramos-Scharrón & MacDonald, 2007).

The high spatial resolution of the SfM-based DSM and orthophoto used in this research provided measurements of sufficient accuracy compared with those derived from other techniques and field-based measurements, representing an effective alternative to ground-based measurements, especially for urban areas where representing complex drainage patterns and microtopographic features is important to accurate mapping of flow paths and A_d.

Topographical Thresholds

The roads in SB showed exceptional vulnerability to gully formation compared with other studies in a range of climates (Figure 5). Topographic thresholds for gully formation observed in SB were similar to gullies observed in Belgium and Portugal (annual precipitation of approximately 700 mm, and silty soils), but low threshold values of S and A_dfor gully initiation in SB indicate greater vulnerability to gully erosion for a given storm event size, compared with thresholds values reported in Colorado (Patton & Schumm, 1975), Portugal, and Belgium (Vandaele et al., 1996) as well as California and Oregon (Montgomery & Dietrich, 1988).

Maugnard, Van Dyck, and Bielders (2014) suggested a potential bias on topographic thresholds using the lower envelope of the S–A_d relationship due to the high sensitivity to outlier values, though the SB point cloud is located generally lower than other point clouds from the literature (Figure 5). This vulnerability is explained in part by the fill materials used to repair gullies, which are typically unconsolidated sand and silt with low critical shear stress and high erodibility that increase gully erosion and sediment yield downstream. Roads also route and focus water flow along straight and steep flow paths, resulting in combinations of S and A_d that form gullies.

Comparing gully erosion studies that use different methods to determine gully formation thresholds is complicated, especially when case studies with different soils, geology, and climates are compared. Rainfall events with a high return period generate higher runoff and increase gully erosion for a given A_d (Vandaele, 1993), although ephemeral gully erosion will start only when a specific interaction of hydrologic, soil properties and management produce overland runoff that exceeds the critical soil shear stress to initiate and sustain gully erosion (Patton & Schumm, 1975; Vandaele et al., 1996). Rossi, Torri, and Santi (2015) noted potential biases in topographic thresholds studies, particularly overestimation of A_d in large watersheds (>30 ha), but this is not the case for the small watersheds of SB.

Urban development also impacts hydrological connectivity and runoff routing in ways that enhance potential for gully formation. Concrete lots, roofs, and parts of unpaved roads with low infiltration capacity generate runoff and route it to the roads, which increase storm water runoff and gully erosion.

Gully erosion rates

Decreasing sediment yield with increasing A_d was observed in SB (Figure 6), which is consistent with a sediment delivery ratio, the fraction of eroded sediment that is transported past the watershed outlet, of less than 1. Walling (1983) indicated that sediment delivery ratiotends to decrease with increased basin size, which has also been reported in other gully erosion studies (Castillo & Gómez, 2016; Poesen et al., 2003; Vandaele et al., 1996).

Figure 6.

Relationship between watershed size and SSL (Specific Soil Loss, the average depth of soil loss in the watershed) from gully erosion in SB, Tijuana, Mexico (circle points and black line). The gray line represent the average relationship reported by Castillo and Gomez (2016) for ephemeral gullies.

The highest SSL estimated in SB (7.4 mm/year) was generally higher than those reported in Castillo and Gómez (2016) for agricultural areas (Table S5 and Figure S1), Oygarden (2003) in Norway (3.7 mm/year), De Santisteban, Casali, and Lopez (2006) in Spain (5.9 mm/year), and Capra, Ferro, Porto, and Scicolone (2012) in Sicily (7.2 mm/year). Cases where gully erosion exceeded values observed in this study were reported by Martínez-Casasnovas et al., 2002in Spain (16.6 mm/year), which was associated with large precipitation events (215 mm over 135 min), and Tebebu et al. (2010) and Zegeye et al. (2016) in subhumid Ethiopia (up to 25 mm/year), which was associated with land use change and poor land management. Urban watershed characteristics such as vegetation removal, impervious surfaces, and hydrological connectivity due to roads can lead to increased gully erosion on unpaved roads. To our knowledge, this is the first attempt to quantify ephemeral gully erosion rates in urban environments at developing countries.

Urban gully erosion on unpaved roads from other studies (i.e., Adediji et al., 2013) was not normalized by time, complicating comparison with observed gully erosion rates in SB. Management practices in SB, especially road maintenance to fill gullies, represent an important contribution to total sediment production in LLCW.

Conclusions

Urbanization has important impacts on soil erosion rates. In the study area, gullies formed almost exclusively on unpaved roads, highlighting them as a major sediment source. Management practices, especially road maintenance that fill gullies with unconsolidated sediment, create an additional and continually replenished source of highly erodible sediment. Lower threshold values of S and A_d for gully incision were found in SB compared with agricultural environments, which is consistent with the high soil erodibility and low critical shear stress measured in the laboratory. Gully erosion rates in Tijuana were higher than almost all of those observed in agricultural watersheds described in the literature. Gully erosion may contribute significantly to the total sediment production, but other processes in the sediment budget need to be quantified for comparison.

The methodology described in this paper can be used in other watersheds to quantify gully erosion on unpaved roads. Our results suggest urgency in implementing management practices such as pavement or other stabilization of dirt roads to mitigate soil erosion. Future studies evaluating the effect of different soil types on gully erosion rates, as well as modeling the effect of road paving on runoff, soil erosion, and sediment loads, are crucial for proper management of sediment in our study area and potentially in urban areas in developing countries.