This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

In the wake of increasing flood disasters, there is an increasing use of flood inundation models to assess risks and impacts at different temporal and spatial scales. Assessing the impacts of extreme climatic rainfall events will require developing design rainfall profiles to represent rainfall under different conditions. Rainfall profiles of different return periods were developed using the Flood Estimation Handbook (FEH) methodology for a small rural catchment of Scotland, to assess flood risks at a catchment scale. Rainfall induced runoff flows were estimated based on a set of catchment characteristics. The channel and floodplain flows were modelled using a two-dimensional hydrodynamic model-TUFLOW. The main channel was represented by a one-dimensional linear channel based on surveyed data and the floodplain topography, was represented by a digital terrain model based on Light Detection and Ranging (LiDAR). A range of hydrological events with different return periods are simulated. Results show that many residential houses and an extensive area of agricultural land are at risk of flooding from extreme events such as a 1 in 100 year flood.

Floods constitute a real and continuing threat to lives, infrastructure and environment throughout much of the world. In recent years the intensity and frequency of flood has been increasing arguably as a result of climate change. The UK is no exception to this. The recent major floods including the catastrophic 2007 summer floods have triggered widespread concern of flood risk and the need for a sustainable approach to flood risk management [

Identification and assessment of flood risk requires modelling of floodplain inundation that allows land owners and river basin managers to make informed decisions on how to manage the risk. Spatially explicit hydrodynamic flood models can play an important role in producing such inundation maps. A key element of these models that make them suitable for flood risk assessment is the ability to provide time-series inundation information about the onset, duration and passing of a flood event [

Advanced deterministic methods of floodplain inundation generally consist of construction of a physically-based fully 2D hydrodynamic model, calibration of the model using historical flood data, use of the best-fit model to simulate synthetic design flood events and interpretation of the model results to generate flood-hazard maps in a Geographical Information System (GIS) environment [

The lack of empirical and real world data to parameterize and validate the output of the models is one of the limitations of the distributed hydrodynamic modelling approach [

Despite the availability of detailed topographic data, there is a lack of long-term observational data especially on river flow discharge in many parts of the rural catchments in Scotland. This may result in the extreme flow events being inaccurately estimated. However, shot-term data, if available, can also be utilized to provide estimates of flow discharge to model flood inundation from larger and less frequent events using data sets from “hydrologically” similar gauged catchments. Application of such techniques in 2D hydrodynamic modelling for ungauged rural river catchments is limited. This paper describes a modelling approach to floodplain mapping where only short-term data set is available on rainfall and flow discharge and there is limited information available on observed flood events. In this study, a fully 2D hydrodynamic model has been applied to develop catchment-scale flood inundation maps from a set of storm events of varying return periods. Furthermore, the model is tested to simulate a past storm event in terms of flood inundation at a local scale.

The Tarland Burn (74 km^{2}), a sub-catchment of the River Dee catchment (2,105 km^{2}) is located in Aberdeenshire, north-east Scotland (

The study area of Tarland Burn catchment (derived from Ordnance Survey 10 m Digital Terrain Model (DTM), reproduced by permission of Ordnance Survey.

The Tarland Burn and its tributaries have been extensively deepened and straightened to drain the surrounding floodplain and wetlands for the benefit of agriculture [

This study was undertaken on a 7 km reach of the Tarland Burn (^{2}. Installed in 1999, the gauging station records water level every 30 s and the values are averaged and stored every 15 min and every hour using a sonic ranging sensor (SR50).

A statistical tool called “WINFAP-FEH” (Version 3) has been used to undertake flood frequency analysis based upon the Flood Estimation Handbook (FEH) method which is recognized as the best practice method for estimating peak flood discharge especially from ungauged catchment. The method essentially involves estimation of the index flood which is the median annual maximum flood (QMED), defining a polling group for the catchment of interest involving hydrologically “similar” catchments, development of a flood growth curve using the pooling group data and derivation of flood frequency curve as the product of QMED and the flood growth curve for a give return period. The QMED represents a typical magnitude of flood which has a return period of two years and is estimated using the following equation based on the use of a set of catchment descriptors essentially for rural catchments [^{0.8510} 0.1536 ^{(1000/SAAR)} FARL ^{3.4451} 0.046 ^{BFIHOST 2}
^{2}), SAAR is the standard average annual rainfall(mm) based on measurements from 1961–1990, FARL is an index of flood attenuation due to reservoirs and lakes and BFIHOST is the base flow index derived from HOST soil data. All of these parameters are readily available and are obtained from the Flood Estimation Handbook (FEH) CD ROM.

As estimating QMED using two years’ flow data generally provides a better estimate than the catchment descriptors method [_{t} = QMED × Z_{T}
_{t}^{3}/s, QMED is median annual maximum flood, m^{3}/s and _{T}

Flood frequency analysis is generally undertaken in two ways. The method of single site analysis is based upon the observed peak flow data series at the target catchment alone and thus not applicable to ungauged catchments. The pooled analysis, on the other hand, is applied to an ungauged catchment based on a set of catchment characteristics to identify a number of gauged catchments (often called donor catchments) which can be considered hydrologically similar to the target catchment. The observed flow data for the donor catchments are then used to estimate peak flows at the ungauged target catchment. For a given distribution, a growth curve for the subject site is estimated as a weighted average of the single-site growth curves for the catchments in the pooling groups and the weightings are a function of the record length and similarity to the subject site [

As the site has a limited rainfall and flow records, design events are estimated using the FEH statistical rainfall runoff method based on a set of catchment descriptors parameters [_{p} (0) = 4.27 DPSBAR − 0.35 PROPWET^{−0.80} DPLBAR ^{−0.54} (1 + URBEXT) − 5.77
_{p} = 2.2 CAREA/T_{p}(t)
_{p}(t)
_{p} (0) is the time to peak of the instantaneous unit hydrograph (hours), Tp (t) is the time to peak of t-hour unit hydrograph (hours), Q_{p} is peak flow of the t-hour unit hydrograph (m^{3}/s), CAREA is the catchment area (km^{2}), t is data interval (hours) and TB is the time base of the unit hydrograph (hours). Similarly, DPSBAR is mean drainage path slope (m/km), PROPWET is proportion of time when the soil moisture deficit (SMD) was equal to or below 6 mm during 1961–1990, DPLBAR is mean drainage path length (km) and URBEXT is extent of urban and suburban land cover and these parameters are usually obtained from the FEH CD-ROM. The design storm duration is estimated using the following equation:
_{p}

The 2D model (TUFLOW, BMT WBM, Build 2011) was applied to the study reach. TUFLOW is a computational hydrodynamic model used to simulate the flow of water along channels and across surfaces. Full details about the model can be found at

The approach adopted in this study includes modelling of the channel as a 1D network nested within the 2D domain representing the floodplain. For this the catchment was divided into a network of small grids and cells. A cell is the smallest unit of model computation. A 5m cell was considered appropriate considering the size of the channel and computational time required to run the model. A ground Digital Elevation Model (DEM) was then developed using the LiDAR data of 1m resolution by transferring elevations to the centre of each cell. This was done by using Vertical Mapper tools available in Map Info (PitneyBowes, Professional Version 10.5.2). The extents of the 2D domain were defined based on the general land topography which mostly included the low-lying floodplain areas that are likely to be flooded. The purpose of defining the 2D domain was to save computational time by excluding permanently dry areas such as hillslopes that would not be flooded. The channel cross sections were taken from ground surveys undertaken between March 2007 and January 2009 and used to define the hydraulic geometry of the main channel. Where the survey data were not available, cross sections were generated from the LiDAR DEM.

The channel roughness or bed resistance values (e.g., Manning’s n) were assigned based on the current land use as suggested in the literatures [

The time series output (such as flow depth and velocity) from the model were processed using the Surface-Water Modelling System (SMS Aquaveo, Version 10.1) software to develop flood inundation maps. The model output can also be imported and viewed in any GIS platform.

The flood frequency curve derived by using the single site analysis based on the Generalised Logistic (GL) frequency distribution (^{3}/s corresponds to a return period of 1 in 25 years. The extended fitting curve indicates a peak flow discharge of 8.2 m^{3}/s and 8.5 m^{3}/s that correspond to approximately 1 in 100 years and 1 in 200 years respectively. There was a good agreement between the observed and FEH catchment descriptors derived estimates of QMED (5.40 m^{3}/s and 5.27 m^{3}/s respectively). The data transfer ratio of 1.024 (

Flood frequency curve at Coull, Tarland derived using single site method.

In the pooling group method, a total of 9 gauging stations (donor catchments) were selected which provided a total of 306 years of flow data. Some stations with data not suitable for either pooling or estimation of QMED were discarded from the pooling group. The flood frequency analysis is based on the GL frequency distribution (

The design flow events of a range of return periods estimated based on a set of catchment descriptor parameters are shown in

Flood frequency curve at Coull, Tarland derived using the pooled method.

Flood discharge estimates for different return periods.

Methods | 1 in 2 years | 1 in 5 years | 1 in 10 years | 1 in 25 years | 1 in 50 years | 1 in 100 years | 1 in 200 years |
---|---|---|---|---|---|---|---|

Single site (m^{3}/s) |
5.4 | 6.3 | 6.8 | 7.4 | 7.8 | 8.2 | 8.5 |

Pooling (m^{3}/s) |
5.4 | 6.9 | 8.0 | 9.7 | 11.2 | 12.7 | 14.6 |

Design flow hydrographs.

Around the southern outskirts of Tarland village the channel overtopped its banks as a result of a storm event in October 2002 affecting many residential properties along the Burnside Road (^{3}/s, which is approximately equivalent to the index flood;

The Burnside Road of Tarland village (

The channel roughness coefficient was found to be the key model parameter. This parameter was adjusted in a bid to match the observed and modelled flood extents around the Burnside Road. It is important to note that the comparison between the observed and modelled flood extents was undertaken by visual examination and judgment as the observed flood extent was very small and localized within a very short stretch of the channel. Hence, no statistical assessment has been undertaken to compare the flood extents. A lumped roughness coefficient of 0.040 resulted in a best fit of the model output with the observed flood event (

At the catchment scale, flood inundation analysis was undertaken using a set of estimated design events of different return period. An example of the flood inundation map developed for an extreme event with a 1 in 100 year probability of occurrence is shown in

Flood inundation map for the October 2002 flood event. At the centre of the circle is the flood inundation around the Burnside Road.

1 in 100 year extreme flood inundation of Tarland.

Flood inundation area under different recurrence interval storm events.

Given the limited range of record lengths at the study site, extrapolation of properties of flow regimes across the homogeneous regions through the regionalization technique can be a source of uncertainty. The UK Flood Estimation Handbook (FEH) recommends that estimating flood event magnitudes from catchment properties and regional climatology should be based on the transfer of analogous data from sites (

The FEH involves the use of an index flood procedure to derive the flood frequency curve at ungauged sites. This procedure is based on the assumption that donor sites have the same flood frequency distribution but differ in terms of the index flood. The index flood is thus taken as a scaling factor for the growth curve which is estimated from catchment characteristics. Flood magnitudes estimated from catchment descriptors are less accurate than flood peak data [

The model simulates flow series representative of the time horizon considered. The flood regime of a catchment is often described through a flood frequency curve. It considers only on a selection of the highest flows which assumes that the flow regime is stationary and the data that have been sampled to construct the flood frequency curves are drawn from a distribution which is not changing. In the absence of reliable climate models, the assumption of stationarity around the time horizon of interest is useful for assessing impacts of future climate on flood risks [

A range of sensitivity tests were performed in order to ascertain how uncertainties in model parameters impact on the robustness of the model output. The key parameters considered in this study were 1D and 2D roughness coefficients. In the case of 1D roughness coefficient, it was determined that a value of 0.040 will adequately represent the roughness condition of the channel mostly covered by small shrubs and vegetation. The inundated areas for a range of roughness coefficients are shown in

Sensitivity analysis of 1D channel roughness coefficient.

The capability of 2D hydraulic models depends on how well the channel flow conveyance and the channel form and bed topography are represented within the model domain. High resolution digital elevation models (DEMs) acquired by airborne laser altimetry (LiDAR) enable the raster-based 2D models to incorporate channel form, bed topography and floodplain geometry more accurately [

Model parameterization and validation is a typical problem of most distributed hydraulic models given due to the paucity of detailed data on flow discharge, channel form and bed topography [

Modelling of flood inundation from extreme rainfall events is a challenge, particularly in many parts of rural catchments of Scotland where there is a lack of long-term observational data, especially on river flow discharge. The study presents the findings from the application of a 2D hydraulic modelling approach to floodplain inundation in a Scottish rural catchment where only limited observation data on flow discharge is available. The FEH flood frequency estimation technique was applied to provide estimates of flow discharge from the extreme events of different probability of occurrences. It was found that there was a good match in the estimates of flow discharge using the single site and the FEH’s pooling group method for the events up to 1 in 5 year return period. For higher and rare events, flow estimates were made using the data set from hydrologically similar gauged catchments applying the techniques of the pooling group method, as it is based on the larger data set of observed events.

The flood inundation modeling using Tuflow indicated that channel roughness is the important model parameter; however, a sensitivity test showed that it is not very sensitive in terms of its impact on the extent of flow inundation. The model was tested using a historical flood event which had a return period of 1 in 2 years. There was a good match between the modelled and the observed flood extents. Nevertheless, there are several sources of uncertainty in the model output. In order to address the uncertainty issues, calibration of the model parameters is required particularly when the long-term gauging records are not available or they are not sufficient to capture the extreme flood events. In such circumstances, modeling of the past events is a useful way to ensure reliability and robustness of the model outcomes.

The model has made use of fine-scale LiDAR data to represent floodplain topography which is increasingly available in the UK. However, the model is tested only for a single observed event. It is recommended that the model should also be tested and evaluated particularly for higher and more intense storm events to enhance the reliability in the model outputs. For this, efforts are being made in the study catchment to capture the extreme events using web-cams which will be useful for validating the model in the future.

The author is grateful to the Aberdeenshire Council, Scotland for providing survey data and other information relevant to the study area.

The author declares no conflict of interest.