This report describes hydrologic models developed by the USGS in cooperation with the City of Brownsville to simulate water budgets (streamflow, evapotranspiration [ET], water-storage volumes, and water diversions and withdrawals) for the nine watersheds in Cameron County, Tex., that routinely contribute runoff to the resacas and other receiving waters in the study area. The models were used to simulate water-budget quantities during January 2022–December 2023. The emphases of the water-budget simulations were (1) to estimate freshwater deliveries from the study area watersheds to the receiving waters of San Martin Lake, the Brownsville Ship Channel, and the Rio Grande and (2) to estimate Rio Grande water diversions to resacas in the study area. This report includes a description of the functionality of the models and the input data, followed by descriptions of the configuration and calibration of the models. Limitations of the models and the model-simulated estimates of the water-budget quantities are also discussed. Simulated estimates of annual outflows of streamflow and diversion inflows for the study area are presented.

The study area consists of 11 selected watersheds (figs. 1 and 2; table 1) in Cameron County, Tex., and includes all of the area that drains to the lower Laguna Madre by way of San Martin Lake and the Brownsville Ship Channel; however, only 9 of the 11 selected watersheds routinely contribute runoff. Most of the study area is in Brownsville, which is the largest city in Cameron County (fig. 1) and had a 2023 population of about 190,000 people (U.S. Census Bureau, 2024).

The climate of the study area is characterized as subtropical subhumid, with hot summers and mild, relatively dry winters (Larkin and Bomar, 1983). The mean monthly temperature ranges from 62.8 degrees Fahrenheit (°F) in January to 87.0 °F in August (National Centers for Environmental Information [NCEI], 2024b). Based on precipitation data collected during 1991–2020, the mean annual precipitation is 25.7 inches (in.), and September is typically the wettest month of the year, with a mean precipitation of 5.73 in. (NCEI, 2024a). In 2022 and 2023, the annual precipitation was 26.18 and 22.15 in., respectively. During 2022–23, daily precipitation exceeded 0.01 in. on 152 days (total number of days from both years) (National Weather Service [NWS], 2024a). The mean interval between daily precipitation amounts equal to or greater than 1.0 in. is about 48 days (Asquith and Roussel, 2004). Although most precipitation occurs in spring, early summer, and fall, amounts greater than 1.0 in. can occur anytime during the year (Larkin and Bomar, 1983). The greatest daily precipitation total during the 2022–23 study period was 4.61 in. on September 20, 2022 (NWS, 2024a).

Tropical storms are relatively frequent in the study area and can generate large amounts of precipitation and extreme storm-tide surge. Hurricane season lasts from June through November, but most tropical storms make landfall or indirectly affect the study area between August and October (NWS, 2024b). The annual mean number of tropical storms is 0.8 (about eight tropical storms every 10 years), and the frequency of occurrence along any 50-mile segment of the Texas coast is one storm about every 6 years (Roth, 2010). During 2022–23, the study area was not subject to tropical storm activity.

The 11 watersheds that compose the study area total about 138,000 acres (about 217 square miles) (fig. 2; table 1). Six of the watersheds are drained mostly by man-made irrigation canals, and five of the watersheds are drained by resacas. The study area includes three major resaca systems: Town Resaca, the upper and lower Resaca de la Palma (also referred to locally as “Resaca de la Guerra”), and the upper and lower Resaca del Rancho Viejo (fig. 2). These systems are intersected and subdivided by roads, railroads, irrigation canals, drainage ditches, and hydraulic control structures (gates and weirs); as a result, the resacas are segmented into a system of individual stream reaches that resemble narrow, linear lakes, ponds, or pools (University of Texas Rio Grande Valley, Texas Water Resources Institute, Texas Commission on Environmental Quality, and Texas Institute for Applied Environmental Research, 2019). The amount of standing or flowing water in these individual waterbodies varies depending on recent precipitation, inflows diverted from the Rio Grande, and hydraulic control structures throughout the resaca network.

In the upper Resaca del Rancho Viejo and upper Resaca de la Palma watersheds (fig. 2), most resaca segments are not sustained by diversions from the Rio Grande and typically resemble marshes or ephemeral lakes. These areas provide for water storage as intermittent pools. The upper Resaca del Rancho Viejo and upper Resaca de la Palma watersheds generally do not contribute runoff in the study area except during or after extreme precipitation events. These upstream watersheds are not included in the watershed model simulations. The resaca segments in the downstream, more developed part of study area remain filled with water by diversions and pumping from the Rio Grande.

The topography of the study area (fig. 3) is relatively flat, with land-surface elevations ranging from approximately 110 feet (ft) above the North American Vertical Datum of 1988 (NAVD 88) in the inland areas to about 10 ft above NAVD 88 closer to the coast (USGS, 2018). Drainage of the study area is eastward towards San Martin Lake and the Brownsville Ship Channel (fig. 1) (University of Texas Rio Grande Valley, Texas Water Resources Institute, Texas Commission on Environmental Quality, and Texas Institute for Applied Environmental Research, 2019). A levee system forms the drainage boundary between the Rio Grande and the study area. Although the Rio Grande is not considered part of the study area (fig. 1), the Rio Grande is the source of the water that is diverted to the study area resacas. Also, some runoff and treated wastewater from the study area are diverted to the Rio Grande. Drainage to the north of the study area flows to the Arroyo Colorado Basin and eventually to the Laguna Madre (fig. 1). The area to the east of the study area consists of low-lying tidal flats and marshes that drain, by way of tidal lakes and marshes, to San Martin Lake and the Brownsville Ship Channel.

The resaca watersheds (where the watersheds drain to resacas) are elevated relative to the canal watersheds (where the watersheds drain to canals) (fig. 3). The alteration of natural floodflows in the Rio Grande from the construction of dams and artificial levees, along with intensive water withdrawals, has contributed to diminished floodflows into these ancient systems. Even before flood control measures on the Rio Grande, these resaca distributary systems, over time, naturally became hydraulically disconnected from their source river through the development of natural levees (Ewing and Gonzalez, 2016). As a further consequence, Town Resaca, lower Resaca del Rancho Viejo, and lower Resaca de la Palma have small drainage areas compared with the adjacent and intervening canal-drained watersheds.

Most of the municipal water supply for the greater Brownsville area is supported by the Rio Grande and the Amistad Reservoir-Falcon Reservoir system (Black and Veatch Corporation, 2020). A supplemental groundwater treatment plant was completed in 2004 as an alternative source of municipal water supply. The plant’s initial (2004) capacity of 7.5 million gallons per day (Mgal/d) was increased to 10 Mgal/d in 2015 (Southmost Regional Water Authority, 2023).

Within the study area, two irrigation districts (Brownsville Irrigation District and Cameron County Irrigation District Number 6) operate infrastructure to deliver water for irrigation. Irrigation district infrastructure (pumps and canals) and water-management operations are independent of the normal drainage and water management of the canal and resaca network that is the subject of this modeling study and report.

The HSPF model software is distributed as part of the Better Assessment Science Integrating Point and Nonpoint Sources (BASINS) system (EPA, 2025). BASINS 4.5 was developed by the EPA (2024b) to support watershed management. BASINS serves as an umbrella-like software package, interfacing with pertinent geodatabases, ancillary datasets, and software programs to facilitate user interaction with HSPF and to help the user understand the hydrological characteristics of a watershed. The HSPF model is one of the primary surface-water modeling components of BASINS. HSPF is an integrated basin-scale model that combines watershed processes with in-stream fate and transport in one-dimensional characterizations of stream channels. The HSPF user’s manual (Bicknell and others, 2001) provides model documentation and underlying model theory. Model parameterization guidance is available in the technical notes provided for the HSPF model (EPA, 2000).

The HSPF model uses a continuous (hourly time step for this study), semilumped-parameter design (Donigian and others, 1995) that can provide continuous water and mass balance by tracking precipitation and streamflow through the conceptual pathways of the hydrologic cycle in a watershed. In HSPF, a watershed is represented by a group of hydrologically distinct areas referred to as “hydrologic response units” (HRUs) that drain to a stream segment, lake, or reservoir; in HSPF, a stream segment, lake, or reservoir is referred to as a “reach reservoir” (RCHRES). The distinct hydrologic characteristics of each HRU are determined on the basis of land use, surficial geography, soil characteristics, and other factors that are deemed to produce similar hydrologic responses to precipitation and potential ET. HRUs are categorized as impervious or pervious land segments, termed impervious land (IMPLND) or pervious land (PERLND), respectively. IMPLND was represented conceptually within HSPF by a surface storage zone, subject to evaporation and runoff processes. PERLND was represented by three interconnected water-storage zones—an upper zone, a lower zone, and a groundwater zone. Each RCHRES was associated with a particular watershed subdivision, referred to as a “catchment,” and receives runoff and sediment from the IMPLNDs and PERLNDs in the catchment. The hydraulics of a RCHRES are simulated by a storage routing method (Donigian and others, 1995).

Figure 4 is a flowchart of HSPF hydrologic processes for IMPLNDs and PERLNDs. The conceptualized movement of overland flow, storage, and evaporation from IMPLNDs is depicted in figure 4A. Overland flow and movement of water through the upper, lower, and groundwater zones of PERLNDs are depicted in figure 4B. The movement of water from IMPLNDs and PERLNDs and between storage zones was controlled by various process-related parameters. Although some parameters are directly measurable, most are determined during model calibration (Martin and others, 2001). The definitions of selected HSPF model process parameters used in the study area are listed in table 2. Complete descriptions of the computational processes and required input model parameters are provided in the HSPF user’s manual (Bicknell and others, 2001).

Unsteady overland flow was routed by using a modified kinematic-wave method (EPA, 2000). The Manning’s and continuity equations (Bicknell and others, 2001) were used with mean overland-flow plane length, slope, and roughness estimates to continuously (at each hourly time step) calculate surface detention storage, from which the overland-flow rate was calculated.

The potential infiltration rate of PERLNDs was computed as an empirical function of soil moisture (fig. 4B). Actual infiltration was calculated from the precipitation excess remaining after subtracting interception losses from precipitation. Precipitation excess was available for surface detention, infiltration, or runoff. Infiltrated moisture can move to three subsurface storage zones—upper zone storage, lower zone storage, and active groundwater storage—or it could be lost as inactive groundwater. The upper zone storage includes storage in surface depressions, surface vegetation, ground litter, and the shallow root zone in the upper few inches of soil. Moisture may leave the surface detention/upper zone storage by ET, overland flow, interflow (shallow subsurface runoff), or percolation to the lower zone. The lower zone storage extends a few feet to the depth of deep-rooted vegetation, which evapotranspires a portion of the moisture stored there. Active groundwater storage sustains base flows in streams during periods of no precipitation. Inactive, or deep, groundwater storage (recharge) does not flow to the stream and was considered lost from the system. No base flow was expected in the study area, as the resaca and canal systems are separated from the Rio Grande. All of the streamflow in the resacas and canals is from diversions from the Rio Grande or from precipitation runoff.

IMPLND was represented by surface storage, evaporation, and overland flow processes (fig. 4A). In an HSPF simulation, water is moved through this network of HRUs and RCHRES drainage areas for each time step specified in the model while conserving water mass—that is, inflow equals outflow plus or minus any change in storage. The processes of surface detention, evaporation, and overland flow on impervious surfaces are modeled in the IMPLND secondary module IWATER in HSPF by functional relations similar to those used for pervious surfaces.

The water budget for the overall model (as well as for individual HRUs and RCHRES drainage areas) can be stated as follows:P + Q_in = ET + Q_out + ΔS,(1)where (all units are in inches)

Individual components can be broken down into subcomponents (for example, waterflow from an HRU [Q_out] is the sum of surface-water flow and interflow). A simplified water-budget equation for the overall model to incorporate these subcomponents results in the following equation with an assumption that the change in storage over time is minimal:P + Q_in ^sw + Q_in ^gw = ET ^sw + ET ^uz + ET ^gw + Q_out ^sw + R,(2)where (all units are in inches)

While maintaining the overall water balance, the model is used to continuously (at each time step) simulate the interaction among subcomponents of the water-budget equation and variations of these subcomponents over time. A conceptualization of the complex hydrologic processes included in the model is depicted in figure 4. The hydrologic processes are described by empirical equations in the model code. Model parameters used in the empirical equations (table 2) are estimated and then adjusted during the calibration of the model (Donigian and others, 1995). Watershed characteristics and typical values and ranges of model parameters (Donigian and others, 1984; EPA, 2000) were used to develop initial (uncalibrated) values for the model parameters.

The model configuration process included delineation of model watersheds, catchments (watershed subdivisions), and stream networks. The study area was delineated into 11 watersheds (fig. 2; table 1) because of the relatively large study area and the major differences in the hydrologic responses of the watersheds. For example, the hydrologic responses of watersheds that are drained by canals are very different from the hydrologic responses of watersheds that include resaca networks, which are typically highly managed through storage of water diverted from the Rio Grande into the resacas. Each watershed except for the upper Resaca del Rancho Viejo and upper Resaca de la Palma watersheds (which were not modeled because they are generally noncontributing) had a separate HSPF model, but some of the models were linked so that model outputs from upstream watersheds provide model inputs for downstream watersheds.

The nine modeled study area watersheds were divided into catchments (table 1). Each catchment represents a modeled stream (resaca or canal) segment and its associated drainage area. The National Hydrography Dataset Plus (NHDPlus) (ver. 2) dataset (EPA, 2012) was used as the foundation for the initial delineation of catchments and stream segments for the nine modeled study area watersheds.

The NHDPlus catchment delineation of the study area initially provided a set of 235 catchments with a mean drainage area of approximately 440 acres. This original set of catchments was further subdivided to produce a more detailed delineation consisting of 307 catchments with a mean drainage area of 337 acres (fig. 2; table 1). The refinement of the NHDPlus catchment delineation was used to (1) define stream reaches with streamflows such that traveltimes through RCHRES more closely approximate the hourly model simulation time step and (2) locate outlets of RCHRES at strategic points such as control structures, streamgages, tributary confluences, and diversion locations. Given the difficulty of using automated techniques for determining catchment boundaries in flat terrain (Al-Muqdadi and Merkel, 2011), refined catchment boundaries were determined with the aid of topographic maps, digital elevation models, and geographic information system (GIS) editing tools. Given the relatively flat topography of the study area (fig. 3), the catchment boundaries may not be exact, but are considered reasonable approximations of catchment configurations within the watersheds and can more accurately characterize timing of runoff and flow routing for the development of the hydrologic models. Also, because HSPF is a semilumped-parameter model (in contrast to a distributed, or grid-based, model) (Maskey, 2022), catchment-scale boundaries are not defined within the model (Brirhet and Benaabidate, 2016). The uncertainty in the accuracy of catchment boundaries was therefore not considered a limitation regarding model representation of the watershed and catchment configuration. Catchment sizes varied within and between watersheds and were mainly dependent on the original resolution of the NHDPlus catchments delineation. For example, the San Martin Lake watershed mean model catchment area is 1,060 acres compared with the Cameron County Ditch 1 watershed mean model catchment area of 233 acres (table 1). Spatial files of the watershed and catchment boundaries are available in Ockerman and Choi (2026).

Using GIS analysis, each catchment was further subdivided into HRUs on the basis of impervious land cover and soil characteristics. Impervious land cover for the study area was determined from the 2019 National Land Cover Database (NLCD) (Multi-Resolution Land Characteristics Consortium [MRLC], 2019). NLCD impervious land cover data represent urban impervious surfaces as a percentage of developed areas for each 30- by 30-meter grid cell in the United States (Dewitz, 2021). The acreage of impervious area within each catchment was represented in the HSPF models as IMPLND HRUs. As shown in figure 5, the more developed (and highly impervious) urban parts of Brownsville are surrounded by relatively large areas of undeveloped and agricultural land with very low imperviousness. Pervious areas are represented in the HSPF models as PERLND HRUs. PERLND HRUs were further characterized according to soil characteristics. Soils data from the U.S. Department of Agriculture [USDA] Web Soil Survey (USDA, 2022, 2023) were downloaded in shapefile format and analyzed within an ArcGIS application (version 10.8.2) (Esri, 2023). Various soil associations (USDA, 2022, 2023) within the study area were grouped and categorized into five general soil types (table 3; fig. 6). The five general soil types were used to characterize the soils in the study area according to their hydrologic response to infiltration, runoff, and ET (USDA, 1977). All soil types represent level to gently sloping terrain. The pervious acreages of each soil type within each model catchment were determined by using ArcGIS in order to define PERLND HRUs. Soil characteristics were used to guide initial estimates for selected HSPF PERLND process-related parameter values.

In HSPF, “stream channel characteristics” for the RCHRES for a given canal or resaca segment are defined in function tables (FTABLES). FTABLES define the relation among surface area, volume, depth, and outflow for each RCHRES. Outflow from a RCHRES was simulated as a function of reach volume or water-surface elevation. Outflow from a RCHRES can also be simulated from a time-series output, which represents, for example, a municipal or agricultural withdrawal or a runoff diversion. Because the canals were mostly dry in 2018, channel geometry properties for canal RCHRES segments were determined from 2018 south Texas light detection and ranging (lidar) data available from the USGS (2018). For selected resaca RCHRES segments, channel geometry data were derived from available bathymetric surveys completed for 10 resacas (Ockerman, 2024). For 79 resaca RCHRES segments without available bathymetric data, FTABLES were scaled from nearby resaca FTABLES that utilized surveyed bathymetric data based on the ratio of unsurveyed to surveyed resaca surface areas.

The development of the HSPF models for the study area required compilation of time-series data that were used for model input and calibration. These time-series data included meteorological, streamflow, and diversion data. Meteorological time-series data (precipitation, water-surface evaporation, and land potential ET) were used as model inputs. Streamflow data (water-surface elevation and discharge) were compared with simulated outputs during the calibration of model process-related parameters. Reported diversions from WWTPs were used to quantify water-quantity inputs to appropriate RCHRES segments.

The primary goal of the hydrologic model calibration was to adjust model process-related parameters so that model-simulated output values, such as streamflow and water-surface elevation, match measured values at streamgages in the study area. In addition, process-related parameter values were determined so that the simulated hydrologic conditions, such as ET and groundwater recharge rates in the study area, were similar to those reported in the literature. The model calibrations were evaluated by following guidelines by Donigian and others (1984), Lumb and others (1994), EPA (2000), Donigian (2002), Moriasi and others (2007), and Duda and others (2012). These guidelines include comparing measured and simulated results such as mean stream discharges of all days, mean of stream discharges of the greatest 10 percent of days, and mean of stream discharges of the lowest 50 percent of days. Additional model-fit statistics (Kittle and others, 1998) were used to characterize the quality of agreement between the measured and simulated hydrologic conditions. Additional model-fit statistics included (1) coefficient of determination (R²) of the linear regression between measured and simulated daily values (stream discharge or water-surface elevation); (2) Nash-Sutcliffe coefficient of model-fit efficiency (NSE) (Nash and Sutcliffe, 1970); and (3) root mean square error. The R² values and NSE describe the variation in the dependent (simulated) variables that was explained by the independent (measured) variable. Calibration quality was assessed using the evaluation criteria in table 8.

Streamflow calibration data (water-surface elevation and stream discharge) were available from five USGS streamgages (map identifiers 3–6 and 8 in fig. 7 and table 4) located in five of the nine watersheds that routinely contribute runoff to the resacas and other receiving waters in the study area and were included in the model for simulation purposes during 2022–23. Depending on when different streamgages first became operational, streamflow data were available starting in February through April 2022 (table 4). For the streamgages in resaca watersheds (Town Resaca and Resaca del Rancho Viejo, map identifiers 3 and 6 in fig. 7 and table 4), calibration was based on comparison of simulated and measured water-surface elevations (figs. 8 and 9). The calibration performances for simulated water-surface elevations at the Town Resaca and Resaca del Rancho Viejo streamgages were measured using 0.3 ft of absolute difference between simulated and measured water-surface elevations (table 8). This calibration criterion was chosen on the basis of a change in water-surface elevation (and resaca depth) that resulted in a range of approximately 6- to 10-percent change in the resaca water-storage volume. For the streamgages in canal watersheds (4, 5, and 8 in fig. 7 and table 4), calibration was based on comparison of simulated and measured streamflow (flow rate, in cubic feet per second) (figs. 10–12). Calibration results for the five stations (comparison of simulated and measured water-surface elevations or discharge) are shown in table 9. Figures 8–12 show hydrographs and scatterplots comparing simulated and measured water-surface elevation or discharge (streamflow) for the five stations.

Calibration data were unavailable for the San Martin Lake, Loma Alta, lower Resaca de la Palma, and Port Brownsville watersheds. Simulation of water budgets for these ungaged watersheds was accomplished by using calibrated model hydrology parameters from HRUs in the five gaged watersheds (Cameron County Ditch 1, North Main Drain, Southmost, lower Resaca del Rancho Viejo, and Town Resaca) for the same HRU types in the four ungaged watersheds.

Calibration results for the Town Resaca watershed were considered very good (table 8), considering both the comparison of simulated and measured water-surface elevations and the R² and NSE model-fit statistics (table 9). Differences between simulated and measured daily water-surface elevations (mean of all days, mean of the highest 10 percent of days, and mean of the lowest 50 percent of days) were within the criterion of 0.3 ft. Model-fit statistics R² and NSE for simulated daily mean water-surface elevations were 0.84 and 0.76, respectively.

Calibration results for the lower Resaca del Rancho Viejo watershed (table 9) were well within the selected criterion of 0.3-ft difference between simulated and measured water-surface elevations (the absolute difference was 0.06 ft for mean water-surface elevations of all days, 0.13 ft for the mean water-surface elevations of the highest 10 percent of days, and 0.01 ft for the mean water-surface elevations of the lowest 50 percent of days). Other model-fit statistics were fair (table 8) (R² = 0.61 and NSE = 0.57; table 9). One factor affecting the R² and NSE statistics was that a substantial part of the streamflow through the Resaca del Rancho Viejo system comes from Rio Grande diversions that are simulated in the model as monthly quantities and divided equally throughout the hourly model time steps. Therefore, the accuracy of the daily water-surface elevation simulation (fig. 9A) was affected by a lack of model input on the actual (hourly or daily) timing of the inflow diversions. Overall, the water-surface elevation simulation is considered satisfactory for water-budget simulations.

Calibration results at USGS streamgage 08474108 North Main Drain at Manzano Street, Brownsville, Tex., for the North Main Drain watershed (table 9) were very good considering the accuracy of daily mean streamflows and other model-fit statistics. The model-fit statistics were considered good (R² = 0.80 and NSE = 0.74). The difference of the daily mean streamflow of all days was 19.5 percent, which is considered fair. The differences of daily mean streamflows of the highest 10 percent of days and the lowest 50 percent of days were less than 5 percent, which is characterized as a very good result (table 8). Overall, the calibration results at this station are considered good (table 8).

Calibration results at USGS streamgage 08474110 Cameron County Ditch 1 at Farm to Market 802, Brownsville, Tex., for the Cameron County Ditch 1 watershed (table 9) were very good for daily mean streamflow of all days and daily mean streamflow of the lowest 50 percent of days (differences of 4.7 and 7.1 percent, respectively). The results for daily mean streamflow of the highest 10 percent of days were fair (difference of 23.0 percent). Results for other model-fit statistics were good (R² = 0.76 and NSE = 0.76). Overall, calibration results for the Cameron County Ditch 1 watershed are considered good (table 8).

For the Southmost watershed, calibration results at USGS streamgage 08474300 Old Main Drain No. 2 at State Highway 4, Brownsville, Tex., were considered very good for daily mean streamflow of all days (−8.2 percent difference; table 9). However, the simulated distribution of high and low streamflows was poor. The difference of daily mean streamflow of the highest 10 percent of days at the streamgage was 30.3 percent, and the difference of daily mean streamflow of the lowest 50 percent of days was −32.3 percent. The model-fit statistics R² and NSE (0.79 and 0.72, respectively) were considered good. Overall, the calibration results for the Southmost watershed are considered fair to good (table 8).

For the five streamgages within the five watersheds that were used for model calibration purposes, simulations of daily mean streamflows or daily mean water-surface elevations were satisfactory (within the range of very good, good, and fair in calibration performance rating, table 8) for all five watershed models. Most of the simulated daily mean water-surface elevation and streamflow values for the highest 10 percent of days and lowest 50 percent of days also were good or very good. Model-fit statistics results for all five watershed models ranged from fair to very good. Calibration results indicate that some of the watershed models (Cameron County Ditch 1 and Southmost) were constrained by the daily mean streamflow accuracy of the highest 10 percent of days and lowest 50 percent of days. However, the overall simulation of stream discharges (streamflow) and water-surface elevations was good to very good for all of the watersheds, indicating that the calibrated watershed models are satisfactory for estimating watershed water budgets.

Calibrated values of selected HSPF process-related parameters were further evaluated by a set of sensitivity analyses to better understand the effects that changes (or possible uncertainty) in selected parameter values would have on simulated runoff. The parameters selected for sensitivity analyses were those with the most effect on infiltration, runoff, and ET, which were the index to infiltration capacity of soil (INFILT), the lower zone nominal storage (LZSN), and the upper zone nominal storage (UZSN). Each sensitivity simulation was made by adjusting a single parameter of the model by relatively large amounts (for example, by decreasing the parameter value by 25 percent) while keeping other model parameters unchanged. Three watersheds were selected for the sensitivity analyses on the basis of watershed characteristics. The Cameron County Ditch 1 watershed is a canal-drained watershed with a moderate amount of development (impervious surfaces account for 39 percent of the total cover [MRLC, 2019; fig. 5]). The Town Resaca watershed contains a relatively high amount of development (impervious surfaces account for 53 percent of the total cover). The Southmost watershed is a canal-drained watershed with a relatively low amount of development (impervious surfaces account for 8 percent of the total cover). For each sensitivity analysis, the change in a parameter value was made for all three watersheds for the entire 2022–23 simulation period. Table 10 provides the results of the sensitivity analyses.

Simulated runoff was relatively insensitive to adjustments of any of the selected parameters—none of the sensitivity simulation scenarios produced changes in runoff greater than 4 percent (table 10). The largest changes in runoff in the sensitivity analyses were caused by increases to LZSN in the Cameron County Ditch 1 and Southmost watersheds (decreases in runoff of 3.0 and 3.4 percent, respectively) and a decrease of UZSN in the Southmost watershed (increase in runoff of 3.7 percent). The Southmost watershed was more sensitive to changes in the parameter values because the three parameters selected for the sensitivity analyses are related to pervious land (PERLNDs) and the Southmost watershed is the least developed watershed in terms of impervious land cover.

Given the results of the model calibrations, as well as the relative insensitivity of simulated runoff to changes in the values of selected parameters, calibrated model hydrology parameters from HRUs in the five gaged watersheds (Cameron County Ditch 1, North Main Drain, Southmost, lower Resaca del Rancho Viejo, and Town Resaca) were applied to the same HRU types in the four ungaged watersheds (San Martin Lake, Loma Alta, Port Brownsville, and lower Resaca de la Palma) to perform water-budget simulations for those watersheds. The resulting ranges of parameter values for all nine study area watersheds included in the HSPF model are listed in table 11. Specific values for each watershed are documented in the model user control input (UCI) files for each watershed (Ockerman and Choi, 2026).

The nine study area watershed models represent a complex hydrologic system described by a set of mathematical equations. Intrinsic to each watershed model is the error and uncertainty associated with the approximations, assumptions, and simplifications that were made when HSPF was used to develop the models. Hydrologic modeling errors typically result from a combination of input data, representation of the physical processes by the algorithms of the model, and parameter estimation during the calibration process (Ely and Kahle, 2004).

A major limitation of the model calibration was the relatively short duration of data (less than 2 years) available from the USGS streamgages in the study area. A longer period of record of streamflow data might facilitate model calibration, model verification, or both over a wider range of hydrologic conditions given that 2022 (watershed area-weighted mean, 26.18 in.) was a typical year and 2023 (watershed area-weighted mean, 22.15 in.) was a drier year based on the mean annual precipitation (1991–2020) of the study area of 25.7 in. (NCEI, 2024a). Also, four of the nine watersheds that routinely contribute runoff to the resacas and other receiving waters in the study area that were used in the model lacked USGS streamgages as a source of calibration data. For the resaca watershed simulations, USGS streamgages in the lower Resaca del Rancho Viejo and Town Resaca watersheds were the only stations available to provide calibration data for simulating water-surface elevations. There were no streamgages in the lower Resaca de la Palma watershed. The streamgages installed near the outlets of their respective watersheds could be valuable for assessing overall water quantities for the resacas. Streamgages in the lower Resaca de la Palma watershed could allow for calibration of model-simulated results. Additional streamgages upstream from the watersheds used in the study could be useful for simulating inflows from the resacas and could also provide information to help characterize the unsteady nature of streamflow routing during runoff events. Three of the six canal-drained watersheds (including San Martin Lake, the largest study area watershed) lacked available streamflow data for model calibration purposes. Model simulations for these ungaged watersheds relied upon use of calibrated parameters from nearby watersheds.

A potentially important source of model uncertainty was the simulation of the operation of control structures (gates and pumps) used by water managers to control runoff during large precipitation events. The watershed models include the capability to simulate these types of management operations. During the time when this report was being written (2024–25), data were not available to quantify the operation of control structures during extreme precipitation and runoff, and the lack of major storm events during the 2022–23 simulation period would have prevented the simulation of any flood-related management actions regardless. However, for future use of the model for simulation of extreme precipitation and runoff, data describing how control structures are operated during large runoff events could improve model accuracy. The effects of control structure operations could be measured by streamgages as well. A streamflow data collection program to measure runoff at selected diversion points, especially during higher streamflow conditions, could provide data for better calibration of management operations within the model.

Quantities and timing of diversions from the Rio Grande into resacas are important model inputs. However, time-series data for these diversions were not available. Estimation of these diversions was treated as part of the calibration process to match simulated resaca volumes with volumes determined from relatively few streamflow data. Therefore, model simulations for periods outside of the original calibration period could require model recalibration, including an estimation of water diversions to resacas. Availability of time-series data for actual measured diversion quantities could enhance the watershed models.

The study area watershed model files are available in Ockerman and Choi (2026). The USGS data release also includes documentation for running the models. Each study area watershed model includes a UCI file and two watershed data management (WDM) files. The UCI file is a text file containing all program input except for time-series data. There is a separate UCI file for all nine study area watershed models.

Model input time-series datasets and simulation output time-series datasets are stored in the WDM files. The WDM files are binary, direct-access data files that can hold several thousand datasets. Time-series data used to drive model simulations, such as meteorological data and streamgage data, are included in one (input) WDM file. Model-simulated time-series data, such as simulated streamflow and water-surface elevations, evaporation and ET, and groundwater recharge, are stored in separate (output) WDM files for each watershed model. Spatial data files of watershed and catchment boundaries also are included (Ockerman and Choi, 2026).

Streamflow discharge from the study area watersheds drains to San Martin Lake, the Brownsville Ship Channel, and the Rio Grande. Streamflow from the San Martin Lake, Loma Alta, lower Resaca del Rancho Viejo, and Cameron County Ditch 1 watersheds flows directly to the drainage area that includes San Martin Lake (fig. 1). Streamflow exiting the Town Resaca, lower Resaca de la Palma, Southmost, and North Main Drain watersheds all combines to exit the study area through the Port Brownsville watershed, which flows to the Brownsville Ship Channel (fig. 1). Some streamflow in the Town Resaca, lower Resaca de la Palma, North Main Drain, and Southmost watersheds drains (mostly through pumping during runoff events) to the Rio Grande. Table 13 provides a compilation of total annual streamflow quantities, in acre-feet, to San Martin Lake, the Brownsville Ship Channel, and the Rio Grande. The Southside WWTP (EPA, 2024d) also discharges treated effluent directly to the Rio Grande; however, the Southside WWTP discharge was not included in the table 13 accounting because it does not flow through a study area watershed and is discharged directly to the Rio Grande.

The northern part of the study area (San Martin, Loma Alta, Cameron County Ditch 1, and lower Resaca del Rancho Viejo watersheds) that drains to San Martin Lake is about 63 percent of the total drainage area for the nine watersheds in Cameron County that routinely contribute runoff to the resacas and other receiving waters (the upper Resaca del Rancho Viejo and upper Resaca de la Palma watersheds generally do not contribute runoff and were excluded from the model simulations) (table 1). About 63 percent of the streamflow exiting the study area flowed to San Martin Lake during the 2022–23 study period (table 13). The Cameron County Ditch 1 watershed was the largest contributor of streamflow to San Martin Lake (table 12). The San Martin Lake and Loma Alta watersheds are large watersheds in the study area, but as explained previously in this section of the report, they include relatively little impervious land cover and generate less runoff yield (runoff per acre) than do the more developed watersheds (fig. 5; table 12). The Loma Alta watershed includes Loma Alta Lake (fig. 2), which under some circumstances can retain substantial quantities of runoff. Also, the San Martin Lake watershed includes sizeable marsh areas (fig. 3) that can store large quantities of runoff.

About 26 percent of the surface-water discharge flowed to the Brownsville Ship Channel from the watersheds in the southern part of the study area (Town Resaca, lower Resaca de la Palma, North Main Drain, Southmost, and Port Brownsville) during the 2022–23 study period (table 13). Although the Southmost watershed is the largest watershed that contributes flow to the Brownsville Ship Channel, most of the study area discharge flowing to the Brownsville Ship Channel originated in the North Main Drain watershed, which includes more impervious land cover than does the larger Southmost watershed (fig. 5). Although the Town Resaca and lower Resaca de la Palma watersheds include considerable impervious land cover (fig. 5) and receive inflows from the Rio Grande, evaporation from resaca water surfaces and management operations to maintain resaca water-surface elevations curtailed the amount of streamflow exiting these watersheds (table 12).

The Rio Grande received about 11 percent of the streamflow exiting the study area (from the Town Resaca, lower Resaca de la Palma, North Main Drain, and Southmost watersheds) during the 2022–23 study period (table 13). Although direct discharge from the Southside WWTP to the Rio Grande was not included in table 13, inflows to the Rio Grande of this treated effluent provide 7,920 acre-feet per year (acre-ft/yr) (EPA, 2024d)—a substantial contribution to streamflow.

Precipitation and runoff in the study area watersheds are typically not sufficient to maintain the water-surface elevations in the resacas where inflows and outflows are managed; inflows to the resacas have been supplemented by diversions from the Rio Grande. Diversions to resacas were estimated as part of the model calibration process to balance water-budget inflows and outflows. Although most of the diversions were to the Town Resaca, lower Resaca del Rancho Viejo, and lower Resaca de la Palma watersheds, some minor diversions also were made to isolated, non-mainstem resacas in the North Main Drain and the Cameron County Ditch 1 watersheds. Table 14 provides estimated annual precipitation and simulated diversion inflows with an estimate of mean resaca surface acreage for selected watersheds. Diversion inflow quantities are provided in watershed inches (inches per acre of watershed area) and acre-feet per year.

Total simulated mean annual Rio Grande diversion to study area resacas during 2022–23 was about 4,000 acre-ft/yr (table 14). The lower Resaca del Rancho Viejo and lower Resaca de la Palma watersheds, with the largest resaca surface areas, required the greatest quantities of Rio Grande diversions during the 2022–23 study period, 1,430 and 1,280 acre-ft/yr, respectively (table 14). These resaca watersheds are similar in size (table 1), amount of impervious area (fig. 5), and amount of resaca surface area (table 14). The Town Resaca watershed has less than half the amount of resaca surface area of the lower Resaca del Rancho Viejo or lower Resaca de la Palma watersheds and therefore required less diversion inflow during the 2022–23 study period (575 acre-ft/yr) (table 14). The North Main Drain and Cameron County Ditch 1 watersheds have smaller resaca surface area and required smaller amounts of diversions during the 2022–23 study period (245 and 418 acre-ft/yr, respectively) (table 14). The diversion quantities, in watershed inches, for the resaca watersheds (Town Resaca, lower Resaca del Rancho Viejo, and lower Resaca de la Palma, which drain to resacas) were considerably greater than those for the canal watersheds (table 14).

Annual precipitation affected diversion quantities, and the amount of precipitation in 2023 was about 4 in. less than it was in 2022 (a decrease of about 15 percent) (table 14). As a result, an increase in diversions of about 44 percent was required in 2023 compared to the diversions made in 2022.