Model Descriptions
HEC-HMS is a semidistributed hydrologic model that is widely used in watershed management and planning because of its ease of implementation, flexible number of hydrologic representations, and low computational demand (
HEC 2015). Prior applications include assessments of proposed reservoirs in decision-making contexts (e.g.,
Emerson et al. 2003;
Ganoulis et al. 2008;
Tingsanchali and Tanmanee 2012;
Robles-Morua et al. 2015). Among the various options, the authors selected the soil moisture accounting scheme for infiltration to carry out continuous simulations in response to Hurricane Alex that also required representations of canopy interception and evapotranspiration. Table
2 describes the hydrologic processes selected in HEC-HMS that were guided by prior applications (
Fleming and Neary 2004;
Robles-Morua et al. 2015) with the aim to be as consistent as possible with the second modeling approach. The tRIBS model is a fully-distributed hydrologic model that uses a triangulated irregular network (TIN) to represent a watershed at multiple resolutions (
Vivoni et al. 2004), as opposed to the subbasin areas used in HEC-HMS. Fully-distributed models capture spatial variations of landscape and meteorological conditions more faithfully, but at the cost of a larger computational demand, and thus with more limitations in decision-making contexts. Several prior applications of tRIBS have focused on flood event simulations (e.g.,
Vivoni et al. 2006;
Nikolopoulus et al. 2011;
Moreno et al. 2013;
Hawkins et al. 2015) and have demonstrated skill in representing the spatially-distributed basin response. Table
2 describes the physical processes represented in tRIBS, including the level-pool reservoir routing (
Mays 2010) developed by the authors to match the formulation in HEC-HMS. Overall, the selection of physical processes in each model captured the continuous soil moisture dynamics influenced by precipitation events and the effects of vegetation on interception and evapotranspiration as well as kinematic wave routing through the stream network and the effects of reservoirs on the flood wave using level-pool routing.
Model Domain, Parameterization, and Calibration
A common set of topographic, soil, and vegetation products were used for setting up and parameterizing the two hydrologic models in the SCW. The watershed domain was delineated from a 30 m digital elevation model (DEM) obtained from the advance spaceborne thermal emission and reflection radiometer (ASTER) [Fig.
2(b)] for both modeling approaches. The authors applied terrain analysis procedures from the HEC-GeoHMS (
HEC 2013) package to derive the basin boundary upstream of the Cadereyta Station and the stream networks for use in both models. A stream cell threshold of
was used to maintain a low number of subbasins in HEC-HMS (a total of 61), as depicted in Fig.
4(a), an enhancement with respect to prior modeling efforts in the SCW where 19 subbasins had been specified (
Ramírez 2010). The hydrographic procedure of Vivoni et al. (
2004) was applied to derive the multiple-resolution TIN domain, resulting in 580,434 Voronoi polygons used in tRIBS as computational elements and an equivalent cell size of 56 m (
Ivanov et al. 2004a;
Vivoni et al. 2005). Fig.
4 shows differences in the domain representation in the two models through the spatial distribution of the terrain slope. The complex terrain of the SCW yields variations in slope that are captured well in tRIBS, whereas only the mean slope in each subbasin is retained in HEC-HMS, leading to large slope overestimates in the upper reaches. While the computational effort for HEC-HMS was possible on a single processor within a desktop computer (Dell Precision T7500, 2.8 GHz Intel processor with 6 GB of RAM), the tRIBS simulations were performed using 16 processors in the ASU Ocotillo computing cluster (2.9 GHz Intel processors with 4 GB of RAM per core) and the application of the parallel processing capabilities described by Vivoni et al. (
2011).
Identical geospatial datasets were used to describe the spatial variations of soil (
ISRIC 2013) and vegetation (
INEGI 1993) properties in each model (Table
1) (see
Cázares-Rodríguez 2016 for spatial maps). Given the low spatial variations in soil types, the authors used the land cover polygons [Fig.
2(c)] as an aggregation scale for specifying landscape properties to both models, following Ivanov et al. (
2004b). As a result, gridded soil parameters from ISRIC (
2013) were averaged within land cover polygons, as were vegetation parameters derived from the moderate resolution imaging spectroradiometer (MODIS) sensor. Cázares-Rodríguez (
2016) further describes the use of pedotransfer functions (
Van Genuchten 1980) and vegetation relations (
Méndez-Barroso et al. 2014) within the model parameterizations. While this procedure reduced the native resolution of the original soil and vegetation products (available at
), it provided a consistent means to input landscape properties to both models. Additional aggregation at the level of subbasins was performed in HEC-HMS, as shown for terrain slope in Fig.
4(a). Table
3 lists the vegetation, soil, and routing parameters, their range of values, and sources used in the HEC-HMS and tRIBS simulations for the SCW. The range of values represents the variations across the land cover polygons within the watershed. The spatial resolution and classification fidelity represented in the models exceed those in previous regional studies (e.g.,
Maqueda et al. 2008;
Ramírez 2010;
Návar 2012).
Model calibration of soil and vegetation parameters (Table
3) followed previous HEC-HMS and tRIBS model applications where values were obtained from literature (L) or remote sensing (RS) products for similar site conditions (e.g.,
Van Genuchten 1980;
Schaap et al. 1999;
Mitchell et al. 2004;
Ivanov et al. 2004a,
b;
Mays 2010;
Robles-Morua et al. 2012,
2015;
Singh and Jain 2015) assumed to be spatially uniform within each land cover polygon and considered to be a reliable means for reducing the overall parameter space to be sampled in the model calibration exercise. As shown in Table
3, the number of soil and vegetation parameter values is large for each model given the spatial variations represented across the land cover polygons, whereas the routing parameter values are spatially uniform within the stream networks depicted in each model. Model initialization was aided by the dry conditions prior to the arrival of Hurricane Alex. As discussed by Vivoni et al. (
2010), semiarid regions in northern México are characterized by ephemeral rivers within alluvial basins with deep groundwater tables. High evaporative demands and low rainfall during the early summer effectively reset the conditions of hydrologic systems to dry states prior to the rainy period occurring later in the summer season. This is in contrast to more humid settings with smaller degrees of seasonality (e.g.,
Nikolopoulos et al. 2011;
Nied et al. 2013;
Massari et al. 2014) where time-variable initial conditions need to be accounted for accurately. As such, a dry state was assumed in each model following prior flood forecasting efforts (e.g.,
Chu and Steinman 2009;
Hawkins et al. 2015).
A manual calibration exercise was conducted with respect to the streamflow record at Cadereyta Station where observations were available at 1–6 h intervals by using the peak error as an objective function given the uncertainties present in other aspects of the streamflow observation and the emphasis of this metric in the stakeholder engagement process. The authors also inspected the streamflow volume of the simulations for the Cadereyta Station during the manual calibration exercise. Simulations were tested in a validation activity against the maximum water levels recorded at the Rompepicos Dam included as a hydraulic infrastructure during the exercise (see Fig.
4 for its location). As such, the model calibration and validation approaches are based on streamflow data at the basin outlet and the maximum reservoir level at an internal location, respectively, exceeding prior efforts in model testing for the extreme event under analysis (
Ramírez 2010), but still focusing attention on the primary concern for stakeholders—the peak discharge—with less emphasis resulting in the other characteristics of the calibrated hydrographs. Model calibration involved varying a limited set of soil and vegetation parameters (labeled C in Table
3) to which the simulated streamflow at the basin outlet was found to be most sensitive through a one-at-a-time analysis using parameter value ranges representing different percentage changes (e.g.,
to
) from nominal values (e.g.,
Forman et al. 2008;
Nikolopoulos et al. 2011) as reported in Cázares-Rodríguez (
2016). Using a similar approach, the authors identified the model parameters that most affected the simulated discharge for this extreme event as
,
, and
for HEC-HMS and
,
, and
for tRIBS (see Table
3 for definitions). In the case of HEC-HMS, calibrated parameters controlled the runoff volume via modification of soil infiltration and storage properties, while for tRIBS the calibrated parameters served a similar purpose (
Ivanov et al. 2004b) and also yielded modifications to the timing of lateral flows. Parameter values that were established well through observations, pedotransfer functions, or remote sensing were not varied during calibration, whereas the sensitive parameters identified earlier could not be derived easily through these methods. This is consistent with the goal of reducing the over-parameterization of semidistributed and fully-distributed hydrologic models which are often criticized for having an excessive number of variable parameters during a calibration process (see discussion in
Fatichi et al. 2016).
Hydraulic Infrastructure Scenarios
Using the calibrated models, the authors evaluated the sensitivity of the flood response in the SCW during Hurricane Alex to the presence of hydraulic infrastructure. Based on stakeholder suggestions during two workshops, four scenarios were considered in a consistent fashion for each model: (1) removal of current hydraulic infrastructure (Rompepicos Dam); (2) evaluation of current conditions (i.e., calibration case); (3) inclusion of an additional large dam at the Entry to the City location; and (4) inclusion of three small detention dams at locations with no current hydraulic infrastructure (i.e., uncontrolled flow). Fig.
4 presents the locations of the proposed hydraulic infrastructure, with the same level-pool routing methodology applied in both models for the scenarios. To include the large dam at the Entry to the City site considered by CERNL (
2010), the authors created a conceptual engineering design for the new dam based on adjusting the elevation-discharge-storage relations of Rompepicos Dam (
Ramírez 2011) using the topographic conditions of the new location and retaining the overall dimensions of Rompepicos Dam. This site is attractive because of its location upstream of Monterrey and its ability to contain both uncontrolled flows and the discharge from Rompepicos Dam. As a lower cost alternative, the three small detention dams were located in key sites where important contributions were simulated to the overall flood response during Hurricane Alex. The authors conceptualized the design of smaller, concrete detention structures with a spillway to apply the level-pool routing at the three sites using local information on topography, channel dimensions, and upstream contributing areas. These sites are intended to collectively reduce and delay the flood contributions from uncontrolled subbasins upstream of Monterrey. Comparisons of the scenarios were carried out for different units in the SCW exhibiting varying terrain properties (labeled Sites I, II, and III) and along various main channel locations (labeled Sites A, B, C, and D) to determine the flood sensitivity in both models.