HYDROLOGICAL PROCESSES Hydrol. Process. 17, 1711– 1732 (2003) Published online 15 May 2003 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/hyp.1270 Floodplain friction parameterization in two-dimensional river flood models using vegetation heights derived from airborne scanning laser altimetry David C. Mason,1 * David M. Cobby,1 Matthew S. Horritt2† and Paul D. Bates2 1 Environmental Systems Science Centre, University of Reading, Reading RG6 6AB, UK 2 School of Geographical Sciences, University of Bristol, Bristol BS8 1SS, UK Abstract: Two-dimensional (2-D) hydraulic models are currently at the forefront of research into river flood inundation prediction. Airborne scanning laser altimetry is an important new data source that can provide such models with spatially distributed floodplain topography together with vegetation heights for parameterization of model friction. The paper investigates how vegetation height data can be used to realize the currently unexploited potential of 2-D flood models to specify a friction factor at each node of the finite element model mesh. The only vegetation attribute required in the estimation of floodplain node friction factors is vegetation height. Different sets of flow resistance equations are used to model channel sediment, short vegetation, and tall and intermediate vegetation. The scheme was tested in a modelling study of a flood event that occurred on the River Severn, UK, in October 1998. A synthetic aperture radar image acquired during the flood provided an observed flood extent against which to validate the predicted extent. The modelled flood extent using variable friction was found to agree with the observed extent almost everywhere within the model domain. The variable-friction model has the considerable advantage that it makes unnecessary the unphysical fitting of floodplain and channel friction factors required in the traditional approach to model calibration. Copyright 2003 John Wiley & Sons, Ltd. KEY WORDS distributed friction; segmentation; synthetic aperture radar; River Severn INTRODUCTION Two-dimensional (2-D) hydraulic models are currently at the forefront of research into river flood inundation prediction. Their popularity has arisen partly because of the inability of one-dimensional (1-D) models to predict certain aspects of out-of-bank flows, and partly because of recent technical advances in two- and three-dimensional hydraulic modelling, for example the development of algorithms to simulate moving flood boundaries (Bates and Hervouet, 1999; Defina, 2000). Given that many flood events of practical significance occur at long reach scales (10–60 km), most models adopt a 2-D finite element approach to minimize the number of nodes required in the computation. These models solve the shallow-water equations at each node of an irregular mesh covering the channel and floodplain. Each node of the mesh must be assigned a topographic height and a (possibly time-varying) friction factor. During a model run, the time-evolving water depths and flow velocities at each node are calculated in such a way as to satisfy the boundary conditions specified. The 2-D nature of these models requires spatially distributed 2-D data for their calibration and validation. Until recently, development of these models was hampered by lack of suitable spatially distributed data. However, remote sensors carried on satellites and aircraft are now proving to be a rich source of such * Correspondence to: David C. Mason, Environmental Systems Science Centre, Harry Pitt Building, 3 Earley Gate, University of Reading, Whiteknights, Reading RG6 6AL, UK. E-mail: [email protected] † Present address: School of Geography, University of Leeds, Leeds LS2 9JT, UK. Copyright 2003 John Wiley & Sons, Ltd. Received 18 April 2002 Accepted 18 November 2002 1712 D. C. MASON ET AL. distributed data. For example, synthetic aperture radar (SAR) satellite imagery of flood inundation extent over large areas can now be used to validate the modelled flood extent (Horritt, 1999, 2001; Horritt et al., 2001). Because of the complex, shallow topographic gradients found on floodplains, reproduction of the inundation extent observed should provide a good test of a 2-D model’s capabilities. Airborne scanning laser altimetry (LiDAR) is a further important new source of data for model parameterization. LiDAR can produce maps of surface height over large areas with a height precision of about š15 cm (depending on the nature of the ground cover) and a spatial resolution of ¾1 m. The technique has found applications in many areas of environmental science, including oceanography (Hwang et al., 2000), coastal mapping (Gutelius et al., 1998), forestry (Naesset, 1997; Magnussen and Boudewyn, 1998), crop height measurement (Davenport et al., 2000) and hydrology (Gomes Pereira and Wicherson, 1999; Marks and Bates, 2000; Bates, 2000). LiDAR data are now routinely collected for UK floodplains by the UK Environment Agency (EA). The ability of LiDAR to provide suitable model bathymetry has been investigated previously (Marks and Bates, 2000; Bates, 2000). LiDAR also shows potential in identifying the river channel, because water generally produces a null return. The channel, together with the LiDAR-derived topography, can be used in the generation of the model domain and mesh. Whilst significant advances have been made in the application of 2-D finite element models to river flood inundation modelling through remote sensing techniques, both in terms of model validation and topographic parameterization, the specification of flow resistance remains a significant problem. Previous work has used a calibration strategy, where friction parameters are adjusted to optimize the fit between model predictions and observations. Owing to computational considerations this generally limits friction parameterizations to those with uniform floodplain friction or a crude representation of the spatially heterogeneous friction surface (Horritt, 2000). Furthermore, owing to the paucity of model validation data sets, the same data are often used for model calibration and validation. This approach is suspect, as shortcomings in model performance can often be compensated for in the calibration process, and so validation is limited by uncertainty in model parameterizations. Thus, there is a clear need to move away from the calibration approach used in previous work (e.g. Bates et al., 1998) to a more scientific method of assigning friction parameters in finite element models of flood hydraulics. Firstly, this will allow models to be validated in a rigorous fashion by reducing the need for calibration, which precludes using inundation extent for validation. Secondly, it will allow the potential spatial complexity of 2-D models to be exploited through the use of distributed roughness values. The scale and spatial complexity of river reaches prohibits the use of traditional techniques (field survey, flume studies) to specify roughness properties; hence, remote sensing techniques are a useful starting point to facilitate such a parameterization. As well as providing topographic information over the floodplain and the position of the channel, LiDAR data can be used to generate maps of vegetation height (Menenti and Ritchie, 1994; Cobby et al., 2001). For floodplains experiencing relatively shallow inundation (<1 m), resistance due to vegetation will dominate the boundary friction term, and so the ability to map vegetation properties such as height over the floodplain is essential to any physically based, spatially distributed approach to friction parameterization. Whilst considerable research has been carried out on flume studies of flow through vegetation (Kouwen and Li, 1980; Kouwen, 1988; Wu et al., 1999; Kouwen and Fathi-Moghadam, 2000), techniques for scaling these results to river reaches of 10–60 km have yet to be developed and the application of the results to numerical models of flood flow has not been attempted. The use of LiDAR to generate distributed friction maps is therefore presented in this paper. The main objectives of the paper are: 1. to investigate how the LiDAR-derived vegetation heights can be converted to friction factors at each node of the model mesh; 2. to compare modelled and observed flood extents using variable and constant floodplain friction factors; 3. to perform sensitivity tests to investigate the effects of different floodplain vegetation scenarios on flood risk. Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) FLOODPLAIN FRICTION PARAMETERISATION 1713 Prior to investigating these, the method used to post-process the LiDAR data in order to extract the floodplain digital elevation model (DEM), vegetation heights and model mesh is outlined. LIDAR SEGMENTATION The classic problem in LiDAR post-processing is how to separate ground hits from surface object hits on vegetation or buildings. Ground hits can be used to construct a DEM of the underlying ground surface, whereas surface object hits, taken in conjunction with nearby ground hits, allow object heights to be determined. A LiDAR range image segmentation system has been developed to perform this separation. The system converts the input height image into two output raster images of surface topography and vegetation height at each point. The river channel and model flood domain extent are also determined, as an aid to constructing the model’s finite element mesh. The segmenter is semi-automatic and requires minimal user intervention. Flood modelling does not require a perfect segmentation as it involves large-area studies over many regions, so a straightforward segmentation technique able to cope with large (approximately gigabyte) images and designed to segment rural scenes has been implemented. Detailed descriptions of the segmenter have been published previously (Mason et al., 1999; Cobby et al., 2000, 2001), and only a summary is given here (Figure 1). The action of the segmenter is illustrated using a 6 ð 6 km2 sub-image from a large 140 km2 LiDAR image of the Severn basin. The data were acquired in June 1999 by the EA using an Optech ALTM1020 LiDAR system measuring time of last return only. At this time of year, leaf canopies were dense, so that relatively few LiDAR pulses penetrated to the ground. The aircraft was flown at approximately 65 m s1 and at 800 m height whilst scanning to a maximum of 19° off-nadir at a frequency of 13 Hz. The laser was pulsed at 5 kHz, which resulted in a mean cross-track point spacing of 3 m. Because current 2-D flood models are capable of predicting inundation extent for river reaches up to 20 km long (Horritt, 2000), with floodplains spanning associated large areas, the segmenter has been designed to use LiDAR data with lower resolution than that used for other applications (Gomes Pereira and Wicherson, 1999). The left-hand side of Figure 2a shows the raw 3 m gridded sub-image used as input to the segmenter. The sub-image contains a 17 km reach of the Severn near Bicton west of Shrewsbury. The bright areas are the overlaps between adjacent LiDAR swaths, which are typically about 550 m wide. The image only appears sensible to the eye when interpolated, as shown on the right-hand side. However, to avoid loss of resolution in interpolation, the raw data are used in the segmentation. The segmenter first identifies water bodies by their zero return (due to specular reflection of the laser pulse away from the sensor), and agglomerates them into connected regions. To remove low-frequency trends, the image is subjected to a detrending step that produces an initial rough estimate of ground heights over the whole image. The detrended height image is then segmented on the basis of its local texture. The texture measure used is the standard deviation of the pixel heights in a small window centred on each pixel. Regions of short vegetation should have low standard deviations, and regions of tall vegetation larger values. The image is thresholded into regions of short, intermediate and tall vegetation, and connected regions of each class are found (Figure 2b). Short vegetation less than 1Ð2 m high includes most agricultural crops and grasses and is a predominant feature on floodplains. Intermediate vegetation includes hedges and shrubs, which may be expected to be significant obstacles to flow over floodplains. Tall ‘vegetation’ greater than 5 m high includes trees and buildings. No attempt is made to distinguish buildings from trees, as there are generally few buildings in rural floodplains such as the one used in this study. An advantage of segmentation is that it allows different topographic and vegetation height extraction algorithms to be used in regions of different cover type. Figure 2c shows the vegetation height map derived from the segmentation. Short vegetation heights are calculated using an empirically derived relationship between the LiDAR standard deviation and measured crop height, which predicts vegetation heights up to 1Ð2 m with an r.m.s. error of 14 cm. In contrast to most LiDAR vegetation height algorithms, this relationship does not rely on a certain fraction of LiDAR pulses being reflected from the ground (Cobby et al., 2001). Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1714 D. C. MASON ET AL. Raw LiDAR data A dense ‘cloud’ of (x, y, z) points. Data re-sampling Heights are sampled onto a 3 m grid. Separate images are constructed using the minimum and maximum heights in each cell. Terrain coverage mask Agglomeration of nearby non-zero pixels identifies continuous land regions. Breaks in this surface are assumed to be water bodies, which are locally connected. Detrending Heights in regions of high standard deviation (e.g. forests) are replaced with a surface connecting the local minima. Then low-frequency first-order slopes are removed using bilinear interpolation in non-overlapping 15 × 15 m windows. Standard deviation of local heights The detrended surface is subtracted from the original, and the standard deviations (s) in 15 × 15 m windows centred on each pixel are calculated. Region identification The standard deviation image is thresholded to separate regions of short (<1.2 m), intermediate and tall (> 5.0 m) vegetation. Disconnected segments of intermediate vegetation (e.g. hedges) are locally reconnected. Each connected region is identified and processed separately in the following stages. Short vegetation regions Vegetation heights (n) are calculated using n = 0.87 ln (s) + 2.57 A fraction(f ) of n is subtracted from the bilinearly interpolated surface (hb) to give the topographic heights beneath the short vegetation (hs), i.e. Tall and intermediate vegetation regions The topography (up to a distance dmax from the regions perimeter) is constructed using an inverse distance (d ) weighting of the surrounding short vegetation topography (hs), and the interior minimum-bilinear surface (hmb). This latter surface only is used for pixels lying beyond dmax from the perimeter. hto = hs dmax −d dmax + hmb d dmax hs = hb - f n The topography is subtracted from the raw data giving the tall and intermediate vegetation Connected river bank locations Topographic heights Vegetation heights Figure 1. Flow chart showing the main stages of the LiDAR segmentation system Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) FLOODPLAIN FRICTION PARAMETERISATION (a) 1715 (b) Raw image Interpolated image (c) (d) Figure 2. (a) Raw 3 m gridded LiDAR data of a 6 ð 6 km2 area in the Severn Basin, UK (left-hand side). The right-hand side shows an interpolated region of the raw data. (b) Connected regions identified in the segmentation. Short vegetation is light brown, hedges are green, tall vegetation dark brown and water regions blue. (c) Vegetation height map. (d) Topographic height map Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1716 D. C. MASON ET AL. The topographic height map (Figure 2d) is constructed in regions of short vegetation by subtracting an empirically determined fraction of the vegetation height from the original data. A comparison with ground control points (GCPs) gave an r.m.s. height error of 17 cm for the topographic surface in regions of short vegetation. This is close to the 10 cm r.m.s. error aimed for in DEMs for model bathymetry (Bates et al., 1997). The figure of 10 cm provides a realistic lower limit for DEM quality, as beyond this the sensor signal becomes indistinguishable from background ‘noise’ generated by microtopography features such as furrows in a ploughed field. In regions of tall and intermediate vegetation, the topographic height map is constructed by interpolation between local minima (assumed to be ground hits) and topographic heights in nearby short vegetation regions. However, ground height accuracy falls off in wooded regions due to poor penetration of the LiDAR through the canopy. Measured over a range of tree canopy densities, the LiDAR-generated topography lay, on average, 64 cm above the corresponding GCPs, with an r.m.s. error of 195 cm. The larger error under trees can be seen in Figure 2d, though fortunately the example floodplain contains few large areas of woodland. Accurate calculation of tree height is well documented, and assumes that pulses reflect from both the tree canopy and the ground beneath (Naesset, 1997; Magnussen and Boudewyn, 1998). Tall and intermediate vegetation heights are derived by subtracting the ground heights from the canopy returns, and are accurate to about 10% of vegetation height. As the LiDAR cannot provide bathymetry within the channel itself, channel bathymetry was generated using 76 channel cross-sections measured at regular intervals along the 17 km reach. Although the alongreach position of each cross-section was known accurately, there was uncertainty in its across-reach position. The best fit of a cross-section with the LiDAR topography was found by cross-correlation, allowing the crosssection to be moved along its length to find the position of maximum correlation. The channel bathymetry was estimated using linear interpolation between adjacent cross-sections, and was then merged with the floodplain topography. The main channel has an average maximum depth of 5 m and a width of 35–50 m. The vegetation height map (Figure 2c) is partly corrupted by bands of increased height in areas of overlap between adjacent LiDAR swaths. These are caused by systematic height errors between adjacent swaths that increase height texture in overlap regions. The problem can be substantially reduced by the removal of heights measured at large scan angles (Cobby et al., 2001). Unfortunately, the correction algorithm requires swaths supplied as individual data files, which were unavailable for this test area. It is hoped that future surveys will provide data in a format suitable for the correction to be applied. MESH GENERATION The flood model chosen for this research was TELEMAC-2D, a 2-D finite element model that has been extensively tested and that solves the continuous shallow-water equations at each time step at a number of discrete spatial nodes within the model domain (Bates et al., 1995). Figure 3 shows a typical triangular finite-element mesh for the reach of the River Severn modelled here. The mesh consists of 5061 nodes and 9962 elements and was generated using the Cheesymesh algorithm (Horritt, 1998, 2000) with the channel and domain boundaries automatically extracted from the LiDAR topography (Figure 2d). A structured grid is used to represent the river channel to ensure that the elements are aligned to model effectively the flow physics and to minimize numerical diffusion. The algorithm inserts a minimum of three elements across the channel to represent the varying bathymetry and fluid velocity. Since the solution accuracy is strongly dependent on element size relative to streamline curvature, it also produces channel elements of uniform angular deviation, with shorter elements around tight meanders (Horritt, 2000). In contrast, the floodplain is represented by a sparser unstructured mesh. This allows a dense mesh to be used in regions of high process gradient to achieve solution accuracy while maintaining computational efficiency by keeping the total number of nodes reasonably small. The only constraints to element size on the floodplain are the channel element size and a parameter to ensure a smooth increase in element size away from the channel. In this paper, no attempt is made to Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1717 FLOODPLAIN FRICTION PARAMETERISATION Figure 3. Finite element mesh. The top inset shows an example polygonal neighbourhood of an interior node (with vegetation class colours from Figure 2b), and the bottom inset shows a polygon on a mesh edge decompose the mesh to reflect topographic or structural features in the floodplain, but future application may have to use meshes developed to reflect hydraulically significant linear features (hedges, ditches and small channels, dykes) found on the floodplain; see Cobby et al. (in press). CALCULATION OF NODE FRICTION FACTORS The potential of 2-D flood models to specify a friction factor at each node of the model mesh is not currently realized. A single value for the floodplain is generally specified, either by look-up tables or via a calibration strategy, whereby the friction parameter is adjusted to gain the best fit between model predictions and observations. LiDAR now offers accurate distributed vegetation height data in floodplains and, therefore, the possibility of improved model performance by using a more physically based friction parameterization (Horritt, 2001). Flow resistance on a floodplain is a combination of the unvegetated surface roughness and that due to vegetation protruding into the flow. As it is assumed that the latter will always dominate (Fathi-Moghadam and Kouwen, 1997), it is the only component considered here. For each mesh node, an instantaneous friction factor must be calculated at each time step, given the frictional material in the neighbourhood of the node and the current water depth and flow velocity there. A node’s neighbourhood is defined as a polygon whose vertices are the centroids of the elements surrounding Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1718 D. C. MASON ET AL. the node (Figure 3). For nodes at the edge of the mesh, two additional vertices are positioned at the bisection points of the lines joining the node with its two neighbouring edge nodes. Sub-regions of connected regions in the vegetation height map are formed by intersecting the node polygon map with the vegetation height map. Each sub-region in a polygon may contain either sediment or one of the three vegetation height classes. If a sub-region contains vegetation then its region’s average vegetation height is attributed to the sub-region. A friction factor is first calculated for each sub-region, prior to constructing an average friction factor for each polygon. Following the approach of Darby (1999), a ‘menu’ approach is adopted whereby a different equation set is used to model flow resistance in a sub-region depending on the type of material in the subregion. For simplicity the menu was constructed to be specific to the test area being studied. Three different sets of flow resistance equations were chosen for the menu, the first modelling sediment in unvegetated channels, the second short vegetation, and the third tall and intermediate vegetation. The only vegetation attribute required by the latter two equation sets is vegetation height. Sediment A channel survey of the River Severn near Montford Bridge (Darby and Thorne, 1996) at the start of the reach revealed a coarse gravel bed (d84 D 88 mm). Following Darby (1999), the approach of Hey (1979) based on 93 field data was used to calculate the Darcy–Weisbach friction factor f: 1 as Rh D 2Ð03 log 3Ð5d f 84 0Ð314 Rh as D 11Ð1 Dmax 1a 1b where as is a dimensionless shape correction factor, Rh is the hydraulic radius and Dmax the maximum depth of flow in the cross-section. Short vegetation Short vegetation on the floodplain includes grasses and cereal crops, which are treated as flexible. When water flows over flexible vegetation, it may bend and reduce in height. When this occurs its boundary roughness may be substantially reduced. The drag force exerted on the water flow by the vegetation depends on its flexural rigidity and stem density. The roughness height is not only related to properties of the vegetation, but also to water flow velocity and depth. Kouwen and Li (1980), in experiments with flow over flexible plastic strips, showed that the deflected vegetation height k is given by 1Ð59 MEI 0Ð25 k D 0Ð14 h 2 h where h (m) is vegetation height (measurable from the LiDAR), (N m2 ) is local boundary shear stress and MEI is the product of stem density M and flexural rigidity in bending J, given by J D EI (where E is the stem modulus of elasticity and I is the stem area’s second moment of inertia). The authors also showed that Equation (2) was valid for a range of natural grasses up to 0Ð9 m high (see table 2 of Kouwen and Li (1980)). It is difficult to measure the individual variables M, E and I because of the heterogeneity of natural vegetation. However, the use of the combined term MEI for natural vegetation has been justified by Kouwen (1988), who pointed out its compensating effect: if the number of stems M per unit area is increased, then this causes a similar change in roughness to a corresponding increase in stiffness EI of individual elements. Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1719 FLOODPLAIN FRICTION PARAMETERISATION Kouwen and Li (1980) calculate the friction factor f from the roughness height k using a semi-logarithmic resistance relation: y 1 n D a C b log 3 k f in which a and b are fitted parameters that are dependent on the ratio of boundary shear velocity Ł (D /1/2 , where is water density) to a critical shear velocity Łcrit , where Łcrit D min0Ð028 C 6Ð33MEI, 0Ð23MEI0Ð106 4 This variation of parameters a and b reflects the fact that, at low shear velocities, natural grasses will bend and behave elastically; however, once the critical shear velocity is exceeded, they will break and lie prone. Values of a and b as a function of Ł /Łcrit are given in table I of Kouwen (1988) (reproduced here as Table 1). Of particular relevance to this study are values for vegetation that remains elastic (i.e. Ł /Łcrit 1), when a D 0Ð15 and b D 1Ð85. The product MEI cannot be measured directly using LiDAR. Of great significance for this study, however, is the fact that MEI has been correlated with vegetation height for a range of growing and dormant grass species in laboratory experiments carried out by Temple (1987) and Kouwen (1988, 2000): 319 h3Ð3 for growing grass 5 MEI D 25Ð4 h2Ð26 for dormant grass Vegetation height h measured from LiDAR can thus be used as a surrogate for the product of the stem density M and flexural rigidity EI. Equation (5) is valid for grasses up to 1Ð0 m high. The correlation between MEI and grass height is very high (95%) for growing grasses, though somewhat less (83%) for dormant grasses. Kouwen (1988) found that MEI values based on grass heights using Equation (5) correlated well with MEI values calculated using the ‘board drop’ test (Eastgate, 1966; Kouwen, 1988). This involves holding a board vertically with its bottom edge on the ground, then dropping it onto a vegetated surface to deflect the stems in a manner similar to flowing water. If the height above ground of the top edge of the board is measured when the board has come to rest, this height can be related to MEI using standard calibration curves (Kouwen, 1988). The method has practical implications, because it may be used to estimate stiffness coefficients for use in Equation (5) for a range of natural grasses. Currently, Equation (5) is valid only for grasses for which it has been calibrated. An extension of these calibrations to estimate MEI values for the diverse range of grasses and cereal crops found on a typical UK floodplain is required. A similar extension is required for Equation (2) determining roughness height k. The above treatment is valid only for submerged grasses, but Kouwen and Unny (1973) point out that the approach can still be used for non-submerged vegetation provided that surface roughness height k is set to the water depth yn . Tall and intermediate vegetation Trees and hedges in the floodplain often have large friction factors and can dissipate a great deal of flow energy, despite occupying a small proportion of the floodplain area. Whereas hedges, grass and cereal crops Table I. Values of a and b (after Kouwen (1988) Reproduced by permission of IAHR Publications) Condition Classification Criteria a b Erect Prone Prone Prone Ł /Łcrit 1Ð0 1Ð0 < Ł /Łcrit 1Ð5 1Ð5 < Ł /Łcrit 2Ð5 2Ð5 < Ł /Łcrit 0Ð15 0Ð20 0Ð28 0Ð29 1Ð85 2Ð70 3Ð08 3Ð50 1 2 3 4 Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1720 D. C. MASON ET AL. may be considered a ubiquitous feature of UK floodplains, stands of forest will generally be found above ground at risk of flooding. However, it is common to find banks of a river lined with trees, a location that may be hydraulically significant. Friction factors for trees and hedges were calculated using the method of Fathi-Moghadam and Kouwen, who investigated flow through non-submerged non-rigid vegetation in the form of coniferous trees (FathiMoghadam and Kouwen, 1997; Kouwen and Fathi-Moghadam, 2000). They showed that the assumption of rigid vegetation can lead to large errors in the estimation of roughness, and that it was necessary to take the flexibility of the foliage into account. The authors extended their original study to include a variety of coniferous trees having a wide range of flexibility, testing small saplings 0Ð3 m high in a flume and full-scale trees 2–4 m high in air. They observed that the measured drag force had a linear relationship with flow velocity, unlike the squared relationship for rigid stems that is often employed. This was due to streamlining of the foliage and, therefore, a reduction in the momentum absorbing area (MAA) of the vegetation as flow velocity increased. A second major effect was due to variation in MAA with flow depth. Plant foliage and stems were regarded as being uniformly distributed in the horizontal plane, but a linear increase of foliage area (and hence MAA) with height was assumed. In verification of this, the observed drag force was found to increase linearly with flow depth. Although these two effects were the dominant ones, there were also secondary effects associated with tree rigidity, shape and MAA varying between different tree species. Stem drag was considered as rigid roughness, but was much less than the drag due to foliage. According to Kouwen and Fathi-Moghadam (2000), the friction factor for the non-submerged condition (where yn /h 1) is 0Ð46 V f D 4Ð06 E yn h 6 where V is flow velocity, E is the tree’s modulus of elasticity, accounts for deformation of the tree as a result of increasing flow velocity and is the density of water. Although our test area contained mainly deciduous trees and hedges, Kouwen and Fathi-Moghadam (2000) point out that while application of Equation (6) to tree species other than coniferous trees may have some error, the estimated friction factor is still expected to be more accurate than available methods that do not consider the flexibility and deflection of trees. Therefore, values of f for all trees and hedges in our test area were assumed to be those for white pine, lying in the mid-range of the coniferous tree species considered, and were calculated using Equation (6), deducing parameters E and from data given by Kouwen and Fathi-Moghadam (2000). Equation (6) assumes that vegetation just covers the ground in plan view, and this assumption is also made for the vegetation within tree and hedge regions in the LiDAR segmentation. Calculation of average friction factor for a node An average instantaneous friction factor is calculated for each mesh node neighbourhood at each model time step based on the friction factors of each sub-region contained in the node. The Darcy–Weisbach friction factor f is linked to the drag force F over an area a of uniform friction for a flow velocity v by F D fav2 /8 7 Considering a floodplain node’s polygon covered by k regions of area ai and mean height hi (i D 1, . . . , k) from different vegetation classes, the total force Fe impeding the flow over the area is the sum of the individual forces Fi due to each vegetation type n Fi 8 Fe D iD1 Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1721 FLOODPLAIN FRICTION PARAMETERISATION Substituting Equation (7) into Equation (8) gives an effective friction factor fe for the node: n fe D fi ai ae 9 iD1 A channel node would contain only one class, namely sediment. Note that Equation (9) considers only current strength and ignores the effect of current direction, e.g. emergent vegetation, such as a hedge, oriented in the direction of the current will have a lower friction factor than one perpendicular to the current due to sheltering effects (Nepf, 1999). MODELLING STUDIES Modelling studies were performed over the Bicton reach of Figure 2 using TELEMAC-2D (Anon, 1998). The reach has a low longitudinal slope of 2 ð 104 (Darby and Thorne, 1996). The modelling studies were based on a 50 year flood event from October 1998. The hydrograph for this event at the gauging station at Montford Bridge at the upstream end of the reach is shown in Figure 4. A RADARSAT SAR image with pixel size 12Ð5 ð 12Ð5 m2 acquired during the flood at 18:00 hours on 30 October 1998 was used to provide an observed flood extent against which to validate the predicted flood extent (Figure 5). The SAR flood extent was extracted using the automatic image processing technique described by Horritt et al. (2001), which uses a statistical active contour model to delineate a flooded region in the SAR image as a region of homogeneous speckle statistics generally having lower backscatter than the unflooded surroundings. Horritt et al. (2001) compared the automatic SAR waterline delineation with that from simultaneously acquired aerial photography, and found that 70% of the SAR waterline coincided with the aerial photography waterline to within 20 m. This is relatively small compared with the average mesh node spacing of ca. 70 m. The main error was due to unflooded vegetation giving similar radar returns to open water. Some error was also due to emergent vegetation at the flood edge. The SAR may place the flood edge at the boundary between totally submerged and beginning-emergent vegetation because emergent vegetation may give higher backscatter than open water due to reflection from the canopy. However, the true flood edge is at the position where the vegetation becomes totally emergent. The RADARSAT flood extent used in this study was manually edited to remove most segmentation errors, as determined by an experienced operator’s interpretation of the SAR image. Model runs were made assuming steady-state conditions over the modelled reach, a reasonable assumption given its short length and the relatively slowly changing hydrograph. The average wave travel time through Figure 4. Hydrograph at Montford Bridge for the flood event considered. The vertical line marks the SAR overpass time at 18:00 hours on 30 October 1998 Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1722 D. C. MASON ET AL. Figure 5. SAR image of the flood event in October 1998. (Reproduced by permission. RADARSAT-1 data Canadian Space Agency/Agence Spatiale Canadienne 1998. Received and processed by DERA, West Freugh, UK. Distributed under licence from RADARSAT International). The darker areas around the river channel show the extent of the flood (lighter indicates larger backscatter) the Bicton reach is about 6–8 h and over such a period around the time of the SAR overpass the flow varies by less than 10%. This is within the error of the gauging station record for a large overbank flood. Although the flow rate was monitored at only the upstream end of the reach, the assumption also allowed a downstream boundary condition to be set. An instantaneous water surface elevation of 51Ð9 m ODN at the downstream end of the reach was estimated by intersecting the flood extent observed there by the SAR with the LiDAR-derived topography. The flow rate at the time of image acquisition at the upstream end of the reach was measured as 308 m3 s1 , providing the upstream boundary condition. A constant viscosity turbulence model was assumed in model runs with a velocity diffusivity of 0Ð1 m2 s1 (Anon, 1998), though results proved fairly insensitive to this parameter. Partially dry elements were detected and the spurious free-surface gradient terms generated in such areas were corrected using the algorithm proposed by Hervouet and Janin (1994). For each mesh node, at each time step, a friction factor f was calculated according to the menu of equation sets described. For short vegetation, at the first time step an initial estimate of boundary shear stress was made using D yn S Copyright 2003 John Wiley & Sons, Ltd. 10 Hydrol. Process. 17, 1711– 1732 (2003) 1723 FLOODPLAIN FRICTION PARAMETERISATION (where (N m3 is the weight density of water and S is bed slope) to obtain the roughness height k using Equation (2). On subsequent time steps was calculated using D fv2 /8 11 where f is the friction factor obtained on the previous time step. For all nodes, the friction factor f was converted to Manning’s n values for use in TELEMAC-2D using the relation yn1/3 12 nD f 8g where g is gravitational acceleration. This is appropriate for uniform 1-D flow in a wide shallow channel, where the hydraulic radius Rh can be assumed equal to normal flow depth yn (Fathi-Moghadam and Kouwen, 1997) and is assumed valid for this case. It was found that, for short vegetation, the values of surface roughness height k calculated from Equation (2) often proved to be greater than the vegetation height h. From the definition of k as deflected vegetation height this should be impossible. This was probably due to the fact that Equation (2) was derived by Kouwen and Li (1980) using slopes often substantially greater than the low slope (2 ð 104 ) within the Bicton reach. As a result, the bed shear stress D yn S is lower for Bicton than for the experimental slopes used, leading to large values of k. Ideally, a version of Equation (2) valid for such low slopes needs to be derived in further flume studies. Such studies should also consider developing versions of Equations (2) and (5) valid for agricultural crops as well as natural grasses. The interim solution adopted here was to limit k to h if Equation (2) predicted a k value greater than h. This probably underestimates friction somewhat, because, from Equation (3), if k is limited then f is also limited. It is known that bankfull discharge at Montford Bridge at the start of the reach is approximately 165 m3 s1 (Darby, 1999). It was assumed that this was also the bankfull discharge for the rest of the reach. An initial steady-state run with this as the upstream boundary condition confirmed that the model did in fact simulate this observed behaviour with no significant floodplain inundation. This confirmed that the channel dimensions and friction factors were approximately correct. The average water surface elevation across the entire reach for this simulation was 53Ð78 m ODN. Figure 6a and b shows floodplain water depths and spatially varying bottom friction for a run assuming steady-state conditions with an input flow rate of 308 m3 s1 (equivalent to the upstream discharge at the time of the SAR overpass). The average water surface elevation was 54Ð79 m ODN, implying an average floodplain water depth of 1Ð01 m. Figure 6b exhibits Manning’s n values varying from 0Ð02–0Ð03 in the channel to up to about 0Ð7 on the floodplain. By comparison with the vegetation height and topography maps of Figure 2c and d, the higher n values occur where water depths are low (e.g. at the flood edge) and/or where vegetation is emergent. Few nodes dominated by short vegetation proved to have bottom shear velocities that exceeded the critical shear velocity of Equation (4), implying that the vegetation was behaving elastically. Figure 6c shows the model and SAR-derived flood extents superimposed on the bottom topography, allowing a qualitative comparison to be made. Table II, however, gives a more rigorous quantitative Table II. Comparison of SAR and model flood edge heights along corresponding sections of waterline, using variable floodplain friction Length of SAR pixels Average slope Model Model SAR SAR Correlation combined (equivalent) perpendicular waterline waterline waterline waterline coefficient sections to waterline average standard average standard (m) height (m) deviation (m) height (m) deviation (m) 2375 190 0Ð040 Copyright 2003 John Wiley & Sons, Ltd. 54Ð649 1Ð51 54Ð651 1Ð79 0Ð89 t0 Probability t > t0 0Ð03 0Ð98 Hydrol. Process. 17, 1711– 1732 (2003) 1724 D. C. MASON ET AL. BOTTOM FRICTION (b) (a) WATER DEPTH (M) 2.000000 1.857143 1.714286 1.571429 1.428571 1.285714 1.142857 1.000000 0.857143 0.714286 0.571429 0.428571 0.285714 0.142857 0.000000 318000 317000 316000 315000 314000 0.763627 0.710582 0.657537 0.604493 0.551448 0.498403 0.445358 0.392314 0.339269 0.286224 0.233179 0.180134 0.127090 0.074045 0.021000 318000 317000 316000 315000 314000 342000 343000 344000 345000 346000 347000 342000 343000 344000 345000 346000 347000 (c) HEIGHT (M) 80.000000 77.247147 318000 74.494293 71.741417 68.988571 66.235718 317000 63.482861 60.730000 57.977139 316000 55.224289 52.471432 49.718571 46.965710 315000 44.212860 41.459999 314000 342000 343000 344000 345000 346000 m3 347000 s1 . Figure 6. Variable-friction model run outputs for an input flow rate of 308 (a) Floodplain water depth (m) (axes are British National Grid coordinates in metres). (b) Bottom friction (Manning’s n). (c) Model (black) and SAR (red) flood extents superimposed on bottom topography comparison of the SAR and model flood edge heights along selected sections of waterline. These sections are in areas where the slope perpendicular to the waterline is as low as possible, the SAR extent is unambiguous, the vegetation is generally short, and the downstream boundary is not close. In these sections (identified by the letters A–E shown in Figure 6c) one might expect SAR-predicted water surface elevation errors due to the approximately one pixel inaccuracy in the determination of the SAR waterline position to be minimized. Each section was 200–740 m in length and contained 16–60 SAR pixels, covering over 2Ð3 km in all. In order to minimize the effects of inaccuracy in the SAR flood edge heights, SAR and model flood edge heights at corresponding positions along the reach were compared using a paired Student t-test in which the data from all five sections were combined into a single statistic for the whole reach. The null hypothesis tested was that Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1725 FLOODPLAIN FRICTION PARAMETERISATION there was no significant difference between the means of the SAR and model waterline heights. Student’s t value for a two-sided paired t-test assuming normally distributed samples is given by 13 t D 1 —2 / [s1 2 C s2 2 2s1 s2 /n] where 1 and 2 are the means of the SAR and model flood edge heights, s1 and s2 are their standard deviations, is the correlation coefficient between corresponding SAR and model heights, and n is the total number of pixels considered over the reach. Use of the paired t-test takes into account the fact that corresponding SAR and model flood edge heights will be correlated due to the gradual drop in height along the reach. However, the height differences at corresponding points are uncorrelated, as is assumed in the test. Table II shows that no significant difference between the SAR and model flood edge mean heights was detectable at the 5% significance level, implying that, for this particular event, the modelled flood extent agrees in most places with the SAR flood extent. Figure 7 shows model and SAR-derived flood extents superimposed on the bottom topography for the parameters of the run of Figure 6, but this time using a temporally and spatially constant Manning’s n value of 0Ð06 everywhere in the floodplain to allow comparison with the variable friction results. The constant value was chosen to be representative of that for a UK floodplain containing a typical mix of grasses, crops, hedges and trees (Chow, 1959; Bates et al., 1998; Horritt and Bates, 2001). Table III compares the SAR and model mean waterline heights for the positions used in Table II, averaged over the reach. Again, no significant difference can be detected at the 5% level. However, the average water surface elevation for the constant-friction model run was 54Ð65 m ODN, which is 0Ð14 m lower than for the variable-friction model run. It is probably unrealistic to expect a height difference in modelled water surface elevations averaged over the whole reach to be reflected exactly in the mean height difference between the constant friction model and HEIGHT (M) 80.000000 318000 77.247147 74.494293 71.741417 68.988571 317000 66.235718 63.482861 60.730000 316000 57.977139 55.224289 52.471432 49.718571 315000 46.965710 44.212860 41.459999 314000 342000 343000 344000 345000 346000 347000 Figure 7. Constant floodplain friction model (black) and SAR flood extents (red) superimposed on bottom topography, for an input flow rate of 308 m3 s1 Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1726 D. C. MASON ET AL. Table III. Comparison of SAR and model flood edge heights along corresponding sections of waterline, using constant floodplain friction (Manning’s n D 0Ð06) Length of SAR pixels Average slope Model Model SAR SAR Correlation combined (equivalent) perpendicular waterline waterline waterline waterline coefficient sections to waterline average standard average standard (m) height (m) deviation (m) height (m) deviation (m) 2375 190 0Ð040 54Ð60 1Ð51 54Ð65 1Ð79 0Ð88 t0 Probability t > t0 0Ð76 0Ð45 SAR flood edge heights sampled over only 7% of the reach waterline. One reason a difference cannot be detected is the relatively high slope perpendicular to the waterline (0Ð04 averaged over all five sections) given the grid cell size of 12Ð5 m onto which the SAR and model heights are mapped. This causes the standard deviations of both mean heights to be relatively high, reducing the sensitivity of the t-test. This is a somewhat common problem for this type of modelling, as images and data tend only to be acquired for large floods at peak discharges when the valley floor is mostly filled. Such data are thus not as discriminatory in terms of hydraulic models as would be the case with a more modest size event. The 0Ð14 m fall in average water surface elevation between variable- and constant-friction model runs is significant in terms of the implied change in flow rate for the floodplain width (up to 500 m) being considered here and for practical studies involved with calculating the overtopping of structures. As, here, we employ a steady-state simulation, this is manifest as a change in model predicted velocity. Figure 8a shows that the variable-friction model run has low floodplain current magnitudes (0Ð1–0Ð4 m s1 that decrease rapidly away from the channel due to higher friction nearer the flood edge. However, the constant-friction model run has higher and more uniform floodplain currents of 0Ð2–0Ð6 m s1 across the floodplain (Figure 8b). It should thus be possible to discriminate between these two model parameterizations using additional validation data from current meters suitably positioned in the floodplain flow. Whilst the relatively high slopes at the waterline have reduced the discriminating power of the SAR data for this particular flood event, the potential of the method can be illustrated using a lower input flow rate. For this floodplain, the latter implies not only lower water depths (resulting in higher friction) but also lower slopes at the waterline, as the flood extent lies within the low-lying floodplain rather than towards adjacent hillslopes and terraces. In this case it is not possible to validate the model waterline heights using the SAR data, but the waterline heights for model runs using variable and constant (Manning’s n D 0Ð06) floodplain friction coefficients can be compared directly. Figure 9 shows the variable and constant floodplain friction model flood extents obtained using an input flow rate of 240 m3 s1 , approximately equivalent to a 5 year flood discharge (Darby, 1999). The average water surface elevations for variable- and constant-friction model runs were 54Ð34 m ODN and 54Ð19 m ODN respectively. The average slope perpendicular to the waterline was 0Ð018, compared with 0Ð040 for an input flow rate of 308 m3 s1 . Differences in the flood extents are now visible in areas of low slope, such as B. The variable and constant floodplain friction mean waterline heights are compared in Table IV to quantify differences seen by eye at the positions identified in Figure 9. A highly significant height difference has been detected. In the above test using variable floodplain friction it has been assumed that the LiDAR data were collected at the same time as the SAR. The LiDAR data were actually collected in June 1998, when there was probably significantly more floodplain vegetation present than at the time of the flood in October 1998. A further model run was carried out using the parameters of the run of Figure 6, including variable friction, but now assuming that short vegetation was 0Ð1 m high everywhere. The average water surface elevation was 54Ð65 m ODN, identical to that of the constant-friction model run. The submerged areas were also similar. If this reduced short vegetation height is correct, the constant floodplain friction model would appear to be a good approximation to the variable-friction model for this particular flood event. Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1727 FLOODPLAIN FRICTION PARAMETERISATION (a) CURRENT (M/S) 0.710000 318000 0.660000 0.610000 0.560000 317000 0.510000 0.460000 0.410000 316000 0.360000 0.310000 0.260000 315000 0.210000 0.160000 0.110000 314000 0.060000 0.010000 342000 343000 344000 345000 346000 347000 (b) CURRENT (M/S) 0.710000 318000 0.660000 0.610000 0.560000 317000 0.510000 0.460000 0.410000 316000 0.360000 0.310000 0.260000 315000 0.210000 0.160000 0.110000 314000 0.060000 0.010000 342000 343000 Figure 8. Floodplain current magnitudes (m Copyright 2003 John Wiley & Sons, Ltd. 344000 s1 345000 346000 m3 347000 s1 for an input flow rate of 308 for (a) variable- and (b) constant-friction model runs. The arrow shows the direction of water flow Hydrol. Process. 17, 1711– 1732 (2003) Copyright 2003 John Wiley & Sons, Ltd. 2560 205 0Ð018 54Ð43 1Ð19 54Ð32 1Ð17 0Ð97 Length of SAR pixels Average slope VariableVariableConstantConstantCorrelation combined (equivalent) perpendicular friction model friction model friction model friction model coefficient to waterline waterline average waterline standard waterline average waterline standard sections (m) height (m) deviation (m) height (m) deviation (m) 6Ð28 t0 <108 Probability t > t0 Table IV. Comparison of variable- and constant-friction model flood edge heights along corresponding sections of waterline for an input flow rate of 240 m3 s1 1728 D. C. MASON ET AL. Hydrol. Process. 17, 1711– 1732 (2003) 1729 FLOODPLAIN FRICTION PARAMETERISATION HEIGHT (M) 80.000000 318000 77.247147 74.494293 71.741417 68.988571 317000 66.235718 63.482861 60.730000 57.977139 316000 55.224289 52.471432 49.718571 315000 46.965710 44.212860 41.459999 314000 342000 343000 344000 345000 346000 347000 Figure 9. Variable (red) and constant (black) floodplain friction model flood extents overlayed on bottom topography, for an input flow rate of 240 m3 s1 Following Darby (1999), the system may also be used to simulate the effects of different scenarios of vegetation cover on floodplain inundation for a given input flow rate. Two scenarios of relevance have been considered for the case of the 50 year flood event. The first involves increasing the height of short vegetation from 0Ð1 to 2Ð0 m while maintaining its extent as before. Floodplain grasses 2Ð0 m high will be largely emergent and flooded on average to a depth of 1 m. This is an extreme case, which is probably outside the bounds of validity of Equation (2) in many areas, but which can serve to provide an upper limit on flood inundation. It was found that, compared with the previous model run with grass 0Ð1 m high, the average water surface elevation with 2Ð0 m high grass was 54Ð82 m ODN, a rise of 0Ð17 m, and the area inundated increased by 3Ð5%, from 259 ha to 268 ha, a relatively small effect. The second scenario investigated the effect of hedges on floodplain inundation. Whereas field crops may change on an annual basis, hedgerows remain part of the underlying skeleton of the landscape and present an (albeit limited) area of relatively high friction to floodwater. The variable-friction model run of Figure 6 was repeated with all the hedges replaced with 0Ð1 m high grass. However, the resulting change in the average water elevation was negligible. This is probably due to the fact that hedges occupy such a small area (5%) of the floodplain. This is not to say that there may not be significant local water level differences across individual hedges, but these cannot be resolved using the current mesh. This question is considered in more detail by Cobby et al. (in press), in which a method of decomposing the mesh to reflect floodplain vegetation structures, such as hedges, having different frictional properties to their immediate surroundings is developed. Large elements in the floodplain are also decomposed if they contain significant topographic features having high height curvatures. This approach makes it possible to predict velocity and depth variations around such features, which may eventually allow the prediction of the subsequent erosion or Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1730 D. C. MASON ET AL. deposition patterns resulting from these variations. This moves flood models closer to realizing the potential of the high data density generated by LiDAR. The approach is compared with the computationally more efficient one used here, of sampling topography and friction values onto relatively large elements in the floodplain. CONCLUSIONS The study has illustrated how vegetation heights can be derived from airborne LiDAR data, and how these heights can subsequently be converted into spatiotemporally varying friction factors at each node of the model mesh in the floodplain. The variable-friction model has the considerable advantage that it obviates the need for the unphysical fitting of floodplain and channel friction factors required in the constant-friction model. This means that it should be possible to use the same model with different input data to predict accurate flood extents for different flood events and different reaches; this is not possible with the constant-friction model, which would, in general, require recalibration. From the modelling studies it can be concluded that, at least for the flood event studied, the method of estimating variable floodplain friction using vegetation heights derived from LiDAR data produces a modelled flood extent that agrees in most places with the extent observed from SAR data. This gives some confidence that the method works, though it obviously needs to be tested for other flood events and on other floodplains of different character to the River Severn. The role of floodplain and bank friction may be much more important for other reaches and events where floodplain flow is more significant, and in those cases the ability to generate distributed friction maps (e.g. Horritt, 2000) based on vegetation heights will represent a significant advance in flood inundation modelling. For this flood event, little difference could be detected between the SAR flood extent and that of the constant floodplain friction model run, implying that the simpler constant floodplain friction model is already a good approximation for this reach. This conclusion is reached irrespective of whether the short vegetation at SAR acquisition time was the same as at LiDAR acquisition time or 0Ð1 m high everywhere. If the former were true, then the 0Ð14 m difference in average water surface elevation between the variable and constant floodplain friction model runs could not, in any case, be detected due to the insufficient resolution of the SAR given the floodplain slopes at the waterline for this event. This uncertainty does highlight the desirability of acquiring LiDAR data on vegetation close to the time of the flood. The potential of SAR for validating model flood extents was illustrated using a lower input flow rate, when, assuming that the constant floodplain friction model extent could be used as a surrogate for the SAR extent, significant differences could be detected between this and the extent predicted using the variable-friction model. This suggests that for other events and topographic settings the method may be more significant, and it is clear that the current practice of only acquiring inundation data for large, valley-filling events is unhelpful in terms of hydraulic model validation. From the simulations performed, it can also be concluded that raising the height of floodplain grasses from 0Ð1 to 2Ð0 m has relatively little effect on water elevation or inundation extent, and that, rather surprisingly, removing the hedgerows has even less effect (at least non-locally). A number of topics for future work may be suggested: (1) The use of an airborne rather than a satellite SAR during a flood event would give much higher spatial resolution and hence improved power for validation. This would also allow the possibility of obtaining a sequence of SAR images during a flood rather than the single sample typically achieved at present. A further advantage of an airborne SAR is that it could be made multi- rather than single-frequency, with polarization information being acquired; this would make possible better flood extent delineation under emergent vegetation, as different frequencies will be scattered differently by the vegetation canopy. An airborne LiDAR system that allowed simultaneous acquisition of multi-spectral scanner data would also be useful for measuring vegetation type and stem density, as well as vegetation height. Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) FLOODPLAIN FRICTION PARAMETERISATION 1731 (2) It is important to carry out further experimental flume studies to measure friction factors for the types of grasses, crops, hedges and deciduous trees found in UK floodplains of low slope, such as that at Bicton, in an extension of the previous work of Kouwen and Li (1980), Kouwen (1988), Kouwen and Fathi-Moghadam (2000) and Wu et al. (1999). In addition, although the empirical friction formulation used here is a good starting point, it can probably be bettered; and the availability of LiDAR data opens the door for further theoretical and experimental work to determine friction as a function of canopy type and height. Such work might also include consideration of the theoretical basis of the roughness parameters employed. There is also a need to understand the contribution of the form friction losses due to microtopography compared with vegetation friction losses, and the mesh decomposition approach of Cobby et al. (in press) may be appropriate here. The main aim of this paper has been to illustrate how LiDAR can be used to estimate a spatiotemporally varying bottom friction, rather than to develop a new friction formulation. (3) The SAR flood extent is currently determined solely on the basis of the satellite image. As a result, SAR waterline heights can vary quite substantially locally along the reach. The usefulness of SAR for validating the model flood extent could be considerably improved by constraining the SAR waterline using the topography and vegetation heights derived from the LiDAR data. As with the model waterline, the height of the SAR waterline should be constrained to change only slowly along a reach. (4) Conditional on obtaining a more discriminating validation data set, one could also explore the extent to which otherwise redundant sub-grid-scale topographic data could be used to enhance the representation of dynamic wetting and drying (Bates and Hervouet, 1999; Defina, 2000; Bates, 2000). The key question here is which topographic features need to be treated explicitly at the grid scale and which, more homogeneous, features can be treated statistically at the sub-grid scale. ACKNOWLEDGEMENTS This work was funded under UK Natural Environment Research Council CONNECT-B grant GR3/CO 030. Thanks are due to the UK Environment Agency for supplying the LiDAR and RADARSAT data, and to Infoterra for CASE student support. REFERENCES Anon. 1998. Telemac Modelling System User Manual Version 4.0 . Direction des Etudes et Recherches, Electricite de France: Paris. Bates PD. 2000. Development and testing of a subgrid-scale model for moving-boundary hydrodynamic problems in shallow water. Hydrological Processes 14: 2073– 2088. Bates PD, Hervouet J-M. 1999. A new method for moving-boundary hydrodynamic problems in shallow water. Proceedings of the Royal Society of London, Series A 455: 3107– 3128. Bates PD, Anderson MG, Hervouet J-M. 1995. An initial comparison of two two-dimensional finite element codes for river flood simulation. Proceedings of the Institution of Civil Engineers, Water, Maritime and Energy 112: 238–248. Bates PD, Horritt MS, Smith CN, Mason DC. 1997. Integrating remote sensing observations of flood hydrology and hydraulic modelling. Hydrological Processes 11: 1777– 1795. Bates PD, Stewart MD, Siggers GB, Smith CN, Hervoeut J-M, Sellin RHJ. 1998. Internal and external validation of a two-dimensional finite element code for river flood simulations. Proceedings of the Institution of Civil Engineers, Water, Maritime and Energy 130: 127– 141. Chow VT. 1959. Open Channel Hydraulics. McGraw-Hill, New York. Cobby DM, Mason DC, Davenport IJ, Horritt MS. 2000. Obtaining accurate maps of topography and vegetation to improve 2-D hydraulic flood models. In Proceedings of EOS/SPIE Symposium on Remote Sensing for Agriculture, Ecosystems, and Hydrology II , Barcelona, 25–29 September, 125–136. Cobby DM, Mason DC, Davenport IJ. 2001. Image processing of airborne scanning laser altimetry for improved river flood modelling. ISPRS Journal of Photogrammetry and Remote Sensing 56(2): 121– 138. Cobby DM, Mason DC, Horritt MS, Bates PD. In press. Two-dimensional hydraulic flood modelling using a finite element mesh decomposed according to vegetation and topographic features derived from airborne scanning laser altimetry. Hydrological Processes DOI: 10.1002/hyp1201. Darby SE. 1999. Effect of riparian vegetation on flow resistance and flood potential. Journal of Hydraulic Engineering, ASCE 125(5): 288–293. Darby SE, Thorne CR. 1996. Predicting stage–discharge curves in channels with bank vegetation. Journal of Hydraulic Engineering, ASCE 122(10): 583–586. Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003) 1732 D. C. MASON ET AL. Davenport IJ, Bradbury RB, Anderson GQA, Hayman GRF, Krebs JR, Mason DC, Wilson JD, Veck NJ. 2000. Improving bird population models using airborne remote sensing. International Journal of Remote Sensing 21: 2705– 2717. Defina A. 2000. Two-dimensional shallow flow equations for partially dry areas. Water Resources Research 36(11): 3251– 3264. Eastgate WI. 1966. Vegetated stabilization of grasses, waterways and dam bywashes. MEng thesis, Department of Civil Engineering, University of Queensland, St Lucia, Queensland, Australia. Fathi-Moghadam M, Kouwen N. 1997. Nonrigid, nonsubmerged, vegetative roughness on floodplains. Journal of Hydraulic Engineering, ASCE 123(1): 51–57. Gomes Pereira LMG, Wicherson RJ. 1999. Suitability of laser data for deriving geographical information: a case study in the context of management of fluvial zones. ISPRS Journal of Photogrammetry and Remote Sensing 54(2– 3): 104–114. Gutelius W, Carter W, Shrestha R, Medvedev E, Gutierrez R, Gibeaut J. 1998. Engineering applications of airborne scanning lasers: reports from the field. Photogrammetric Engineering and Remote Sensing LXIV(4): 246– 253. Hervouet J-M, Janin J-M. 1994. Finite element algorithms for modelling flood propagation. Modelling Flood Propagation Over Initially Dry Areas, In Molinaro P, Natale L (eds). American Society of Civil Engineers: New York; 102–113. Hey RD. 1979. Flow resistance in gravel-bed rivers. Journal of the Hydraulics Division, ASCE 105(4): 365– 379. Horritt MS. 1998. Enhanced flood flow modelling using remote sensing techniques. PhD thesis, University of Reading, UK. Horritt MS. 1999. A statistical active contour model for SAR image segmentation. Image and Vision Computing 17: 213–224. Horritt MS. 2000. Development of physically based meshes for two-dimensional models of meandering channel flow. International Journal of Numerical Methods in Engineering 47: 2109– 2137. Horritt MS. 2001. Calibration of a 2-dimensional finite element flood flow model using satellite radar imagery. Water Resources Research 36(11): 3279– 3291. Horritt MS, Bates PD. 2001. Predicting floodplain inundation: raster-based modelling versus the finite element approach. Hydrological Processes 15: 825– 842. Horritt MS, Mason DC, Luckman AJ. 2001. Flood boundary delineation from synthetic aperture radar imagery using a statistical active contour model. International Journal of Remote Sensing 22(13): 2489– 2507. Hwang PA, Krabill WB, Wright W, Swift RN, Walsh EJ. 2000. Airborne scanning lidar measurement of ocean waves. Remote Sensing of Environment 73: 236– 246. Kouwen N. 1988. Field estimation of the biomechanical properties of grass. Journal of Hydraulics Engineering, ASCE 26(5): 559– 568. Kouwen N. 2000. Closure of ‘Effect of riparian vegetation on flow resistance and flood potential’. Journal of Hydraulics Engineering, ASCE 126(12): 954. Kouwen N, Fathi-Moghadam M. 2000. Friction factors for coniferous trees along rivers. Journal of Hydraulics Engineering, ASCE 126(1): 732–740. Kouwen N, Li RM. 1980. Biomechanics of vegetative channel linings. Journal of the Hydraulics Division, ASCE 106(6): 713– 728. Kouwen N, Unny TE. 1973. Flexible roughness in open channels. Journal of the Hydraulics Division, ASCE 99(5): 1085– 1103. Magnussen S, Boudewyn P. 1998. Derivations of stand heights from airborne laser scanner data with canopy-based quantile estimators. Canadian Journal of Forest Research 28: 1016– 1031. Marks K, Bates PD. 2000. Integration of high-resolution topographic data with floodplain flow models. Hydrological Processes 14: 2109– 2122. Mason DC, Cobby DM, Davenport IJ. 1999. Image processing of airborne scanning laser altimetry for some environmental applications. In Proceedings of EOS/SPIE Symposium on Image Processing for Remote Sensing V , vol. 3871, Florence, 20–24 September; 55–62. Menenti M, Ritchie JC. 1994. Estimation of effective aerodynamic roughness of Walnut Gulch watershed with laser altimeter measurements. Water Resources Research 30(5): 1329– 1337. Naesset E. 1997. Determination of mean tree height of forest stands using airborne laser scanner data. ISPRS Journal of Photogrammetry and Remote Sensing 52: 49– 56. Nepf HM. 1999. Drag, turbulence and diffusion in flow through emergent vegetation. Water Resources Research 35(2): 479–489. Temple DM. 1987. Closure of ‘Velocity distribution coefficients for grass-lined channels’. Journal of Hydraulics Engineering, ASCE 113(9): 1224– 1226. Wu F-C, Shen HS, Chou Y-J. 1999. Variation of roughness coefficients for unsubmerged and submerged vegetation. Journal of Hydraulics Engineering, ASCE 125(9): 934– 942. Copyright 2003 John Wiley & Sons, Ltd. Hydrol. Process. 17, 1711– 1732 (2003)
© Copyright 2026 Paperzz