Application of a rainfall-runoff model for regional-scale flood inundation mapping for the Langat River Basin

Floods are among the most destructive events that occur, causing huge losses in terms of life and economics. Many studies have been carried out to forecast flood probability and predict losses. Flood forecasting is mainly influenced by hydrologic data, the availability of which are subject to various unpredictable factors including antecedent conditions and precipitation. Significant improvements are needed in the systems used to predict the floods. With the increasing magnitude and frequency of extreme hydrological events–e.g., droughts and floods–it is important to develop models capable of estimating floods reasonably accurately using the data available. When urban development spreads onto flood plains, this reduces floodwater storage and diversion routes. The impact of urbanization includes alteration of a watershed's response to rainfall, increasing volumes and peak flows, and flood risk downstream, reducing low flows and increasing pollution (Brilly Rusjan & Vidmar 2006).

Prata et al. (2011) focused on the application of HEC-RAS to analyzing river behavior and studied the Taquaracu River in the City of Ibiracu, in Brazil using maximum flows derived from rainfall data. HEC-RAS was used with these to compute water levels along the river and its tributaries under different flow conditions. Merwade et al. (2008) proposed the use of GIS techniques for creating river terrain models. These were first applied and then cross-validated using data for three study reaches: the Brazos River in Texas, the Kootenai River in Montana, and Strouds Creek in North Carolina. Solaimani (2009) integrated GIS and HEC-RAS to use hydraulic analysis to separate the high- and low- risk areas on the floodplain. The results from HEC-RAS were displayed using GIS.

To predict floods, accurate runoff data are required. One computer model that has become increasingly popular is the Kinematic Runoff and Erosion Model, KINEROS2 (K2). In addition to K2, the Automated Geospatial Watershed Assessment (AGWA) tool is a versatile means of hydrologic analysis that: (1) provides a simple, direct and repeatable method for hydrologic modeling; (2) makes it possible to use GIS databases; (3) is compatible with other geospatial basin analysis software environments (Mirzaei et al. 2014); and (4) is useful for developing alternative scenarios and future simulation works at multiple scales (Miller et al. 2002; Goodarzi et al. 2012; Mirzaei et al. 2013). AGWA provides the functionality for modeling and evaluation using the Soil and Water Assessment Tool and K2. Nedkov & Burkhard (2011) reiterated the modeling process in the AGWA GIS environment with five main steps–watershed delineation and discretization, vegetation cover and soil parameters, precipitation writing files, running the model, and visualizing the results.

The main objective of this study was to develop a model to delineate the flood plain. The idea behind it was to use a model to analyze the Langat River Basin's behavior under different conditions using advanced technologies. Within this objective, the aims of the study were to: analyze the frequency analysis maximum rate of rainfall annually for different durations, simulate different water levels and flow velocities corresponding to different return periods from the available database using HEC-GeoRAS and HEC-RAS, and analyze the water level along the Langat River and its tributaries, under different flow conditions.

The main purpose of frequency analysis is to estimate the parameter vector

The geometry set-up in HEC-GeoRAS involves digitizing the stream centerline, predicting overbank flowlines, levees, and providing polygon coverage delineating surface roughness. This was facilitated by visual interpretation of the Digital Terrain Model, and comparison of channel and floodplain cover, with estimates of the channel's roughness characteristics using Hicks & Mason (1998) method and the floodplain. Finally, very closely spaced cross-sections were digitized, ensuring that these were kept as perpendicular to the line of flow as possible. Because the primary channel meanders across the active gravel channel, changing direction frequently, the placement of section lines takes considerable effort as HEC-RAS requires that cross-sections do not overlap anywhere. Channel and cross-section lines are then encoded with z values from a TIN surface. Finally, pre-processed geometry data were exported to HEC-RAS. (Howard et al. 2012)

HEC-RAS was developed for one-dimensional and unsteady flow hydraulic calculations, and sediment transport modeling (Brunner 2006). Under steady flow conditions, HEC-RAS calculates water surface profiles based on the input geometry, the gradient, and Manning's n for any number of discharge rates. Water surface profiles referenced in the software are equivalent to trim-lines, wash limits, or peak-flow water heights. The system is capable of modeling channel networks or a single reach, and provides modeling of subcritical, supercritical, and mixed–i.e., a mixture of subcritical, critical, and supercritical flow–flow regimes (Brunner 2006). HEC-RAS calculates water surface profiles for successive channel cross-sections by solving the energy equation using the standard step-backwater method. Conveyance, channel velocity, and energy loss are accounted for by Manning's equation. To use HEC-RAS effectively for steady flow in natural channels several assumptions are made: (1) flow is steady, (2) flow varies gradually along the reach, (3) flow is one-dimensional, and (4) the channel slope is low (<10%) (Brunner 2001; Howard et al. 2012).

The predicted extent of inundation was derived by exporting the HEC-RAS output water surface to HEC-GeoRAS, where the water surface was overlaid, showing excellent correspondence with the flood inundation delineated. Once in GeoRAS, inundation data for each modeled scenario were converted to TINs and then GRID surfaces. Velocity data were exported to GeoRAS in point format, each point representing the estimated velocity at a cross-section. These were also interpolated to a TIN, and the vertices converted to a GRID form and overlaid onto their respective terrain surfaces for validity checking. HEC-RAS includes tools that can be used to simulate raising or lowering the bed level, and to simulate damming. These were explored and evaluated, to determine whether they were more or less effective than the terrain modeling approach when using high resolution (z) geometry (Howard et al. 2012).

The KINEROS rainfall-runoff-erosion model was developed in the 1970s and has continued to evolve and improve (Woolhiser et al. 1990; Goodrich et al. 2006). It is now known as KINEROS2 (K2). It is a physical-event based, distributed and dynamic model that predicts surface runoff, erosion losses, infiltration amount and interception depth from the watershed, arising predominantly from overland flow. The watershed is approximated by a cascade of overland flow planes, channels and impoundments. The flow planes can be split into multiple components with different slopes, roughness, soils, etc., and contiguous planes can have different widths. In an overland flow conceptual model, small-scale spatial distribution of infiltration variability can be represented and parameterized for numerical efficiency, and the micro-topography is inserted into the simulation. In most urban element models, the runoff is based on the pervious and impervious fractions. In K2, however, infiltration is dynamic, and interacts with both rainfall and runoff. The conceptual infiltration model incorporates two layers in the soil profile, and soil moisture is redistributed during any hiatus in the storm.

Sensitivity of the model to change was based on a selected set of parameters. By modifying them from their initial value, the degree of change in peak runoff was determined for each event, so that those sensitive to changes in peak runoff were identified. The parameters evaluated in this way were–saturated hydraulic conductivity (Ks_Ch and Ks_Ul), Manning's n (n_Ch and n_Ul), mean capillary drive (G_Ch and G_Ul), all for channels and uplands respectively, and coefficient of variation of Ks (CV_Ks), upland interception (I_Ul), and rain-splash coefficient (C_f).

The Langat River modeling resulted in the creation of flood extent and hazard, water depth and flow velocity maps. These were analyzed to derive explanations for various scenarios.

A GEV distribution was fitted separately to the running annual maxima for durations of 15, 30, 60, 90 and 120 minutes, for time-series data. Table 1 shows the estimated GEV parameters μ, γ and К. As expected μ increases with increasing duration.

Table 1

Estimated GEV parameters for D = 15, 30, 60, 90 and 120 min

D (min) .	.	.	.
15	28.676	0.097	0.2441
30	47.129	0.223	0.1562
60	67.099	0.228	0.1896
90	75.676	0.228	0.1519
120	78.088	0.255	0.1743

D (min) .	.	.	.
15	28.676	0.097	0.2441
30	47.129	0.223	0.1562
60	67.099	0.228	0.1896
90	75.676	0.228	0.1519
120	78.088	0.255	0.1743

View Large

The discharge data were obtained from K2, which simulates the hydrologic elements of a meteorological event like peak flow and infiltration (Nikolova et al. 2009; Mirzaei et al. 2013). As runoff data are available for the basin outlet, the model was calibrated and validated on the basis of three events there. The NSE calculated for the calibrated model was 0.88 and the model was validated using data from storm events in the same year as those used for calibration. The NSE for this event was 0.74, which is reasonably good.

Nineteen scenarios were studied using different combinations of return period and duration. The model results show flood depths of up to 18 m (return period 100 years, duration 120 minutes). The greatest flood depth occurred upstream on the main channel, in a largely undeveloped area–perhaps because of the rapid discharge to the main channel there.

Peak flows on the Langat River have been forecast with respect to rainfall events of different durations and return periods using the K2 model. The maximum peak flows related to different return period storms of different durations were also obtained.

Different water levels corresponding to different return period/duration storms were simulated using HEC-RAS. The maximum flood depth for the 100-year return period event of 120-minute duration is 18 m, while for the 15-minute 5-year return period event, it is 11 m.

Flow velocities corresponding to different storm return periods and durations were also simulated with HEC-RAS. For the 100-year return period storm of 120-minutes duration, the maximum velocity is 9.11 m/s, while for the 5-year return period 15-minute event, the maximum velocity is 2.5 m/s.

Using ArcGIS, it is possible to concentrate on the hydraulic model rather than preparing the data. Incorporating ArcGIS in the modeling can improve the accuracy of forecasting and save costs, subject to the quality of the TIN map.

Flood inundation prediction under different probabilistic scenarios can be used as a basis for floodplain risk management, as well as planning agricultural, industrial and urban expansion. It is important to locate new developments in areas of lower flood risk along the rivers, to minimize the potential social and economic impacts of flood hazard.