**1. Introduction**

Erosion, defined as the detachment, transport, and spatial redistribution of soil particles [1,2], contributes to environmental degradation around the globe [3]. Urbanization can lead to an increase

in erosion and the discharge of terrigenous materials into downstream ecosystems, including inland lakes, reservoirs, estuaries, and oceans.

Sheetwash, rill, and gully erosion, hereafter referred to as hillslope erosion, are frequently associated with anthropogenic soil disturbance and are often related to land use change, such as deforestation and urban development [4,5]. Hillslope erosion processes have been well characterized in agricultural settings, but not in urbanized areas where high erosion rates have also been reported [6]. Hillslope erosion rates typically decrease as bare soil in construction sites is replaced by impervious surface and vegetation [7]. Conversely, in developing countries, soil exposure such as vacant lots and unpaved roads can persist for longer periods [8], which increases hillslope erosion rates compared to other urban watersheds with high impervious cover fractions [9] and storm-water management practices [10].

Hillslope erosion and sediment production can be simulated using numerical models that consider the relationship between terrain attributes and climate regimes [11,12]. These models vary in structure, assumptions, and data requirements [13,14]. Erosion modeling is often used to simulate various erosional processes, such as sheet, rill, gully, and channel erosion, to develop sediment budgets and to assess the effect of Best Management Practices (BMPs) on total sediment reduction. The sediment budget is the quantitative tracking of contributing sources, sinks, and spatial redistribution of sediments over a given time scale [15].

Several soil erosion studies have focused on sheet and rill processes [16–19], but ephemeral gullies can also contribute a significant source of sediment at the catchment scale [20–22], especially in arid and semi-arid areas [23]. Such gullies are caused by concentrated overland flow [24] and are commonly cleared by tillage operations [25] or, in urban environments, filled with unconsolidated sediment during grading [9]. These erosional features form due to a complex relationship between terrain and management characteristics such as slope, land cover, soil properties, climate regime, and management activities [26].

The Annualized Agricultural Non-Point Source (AnnAGNPS) model is a simulation tool developed by the USDA-Agricultural Research Service and the Natural Resources Conservation Service (NRCS) to evaluate the effect of land use and management activities on watershed hydrology and sediment transport [12]. AnnAGNPS simulates runoff and sediment generation by tracking their transport through the channel network (AnnAGNPS reaches) at the watershed scale on a daily time step. AnnAGNPS simulates different erosional processes (i.e., sheet, rill, and gullies) as well as channel sources. Sheet and rill erosion are simulated using the Revised Universal Soil Loss Equation (RUSLE). Ephemeral gully erosion is simulated using EGEM (Ephemeral Gully Erosion Model) whose hydrologic routines to calculate peak and total discharge are estimated following the SCS curve number (CN) methodology [27], and gully width and soil erosion calculations are based on the Chemical, Runoff, and Erosion from the Agricultural Management Systems (CREAMS) model [28]. The model simulates colluvial storage of sediment using the Hydro-geomorphic Universal Soil Loss Equation (HUSLE), which calculates a delivery ratio based on particle size distribution and flow transport capacity [29].

The AnnAGNPS model has been tested in small Mediterranean watersheds (< 1.3 km2) [14,23,30]. Licciardello et al. [30] evaluated AnnAGNPS in a steep catchment under pasture in Eastern Italy. Taguas et al. [23] evaluated the effect of different management activities on total sediment reduction in an agricultural environment in Spain, where ephemeral gullies are a significant contributor to the total sediment production. Gudino-Elizondo et al. [14] reported good performance of AnnAGNPS in simulating ephemeral gullies at the neighborhood scale in an urban watershed. However, AnnAGNPS has not been tested to model hillslope erosion rates in an urbanizing catchment under different soil types and land uses.

This study aims to (1) test the capabilities of AnnAGNPS to simulate runoff and sediment production in an urban watershed in a developing country context and (2) use the model to constrain the sediment budget in order to inform management and policy designed to mitigate sediment loads downstream. This paper addresses the following research questions: (a) How accurately does AnnAGNPS simulate water and sediment loads in an urban watershed in a developing-country context where ephemeral gullies are likely to be a significant source of sediment? (b) What processes generate sediment in the watershed, and what is the role of soil properties and land use? (c) How does storm size affect the sediment load from different hillslope processes (sheet and rill, and gully erosion)? (d) Where are hot spots of sediment production, and what watershed characteristics control sediment production? and (e) What are the implications of the sediment budget and distribution of hotspots for management designed to mitigate sediment loads?
