The topographic signature of vegetation ... - Wiley Online Library

WATER RESOURCES RESEARCH, VOL. 47, W06525, doi:10.1029/2010WR010319, 2011

The topographic signature of vegetation development along a braided river: Results of a combined analysis of airborne lidar, color air photographs, and ground measurements W. Bertoldi,1 A. M. Gurnell,1 and N. A. Drake2 Received 8 December 2010; revised 22 April 2011; accepted 3 May 2011; published 29 June 2011.

[1] This paper combines archived remotely sensed data (airborne lidar and digital color air

photographs) with nonsynchronous ground observations (including observations of topographic form and vegetation cover and growth) to test the hypothesis that colonization of exposed river sediments by riparian trees has an impact on channel form and to quantify any impact that is identified. This is achieved along a 21 km reach of the braided, gravel bed Tagliamento River, in northeast Italy, where the width of the braided corridor typically exceeds 800 m. Lidar data are analyzed to extract a 2 m resolution digital evolution model (DEM) and determine riparian vegetation extent, height, and structure within the active corridor. Aerial photographs are used to map the topography of the submerged parts of the corridor. These data are divided into 1 km length subreaches, which possess strong contrasts in vegetation height and extent. Joint analysis of vegetation and morphological properties of these subreaches reveals significant associations between vegetation properties and reach morphology. Residuals from a gamma function fitted to the topographic data for each subreach show a good fit with poorly vegetated reaches, but a weakening fit with increasing vegetation cover, largely as a result of the appearance of secondary peaks in the elevation frequency distribution associated with the heavily vegetated areas. Furthermore, the overall skewness and kurtosis of the elevation frequency distribution within each of the subreaches are both significantly correlated with vegetation extent, height, median elevation, and growth rate, indicating a clear topographic signature of vegetation development along this braided river that reflects sediment accumulation within and around the vegetated patches.

Citation: Bertoldi, W., A. M. Gurnell, and N. A. Drake (2011), The topographic signature of vegetation development along a braided river: Results of a combined analysis of airborne lidar, color air photographs, and ground measurements, Water Resour. Res., 47, W06525, doi:10.1029/2010WR010319.

1.

Introduction

[2] Over the last two decades, field investigations have supported evolving concepts of two-way interactions between vegetation and fluvial processes within river corridors [Gurnell et al., 2001, 2005; Corenblit et al., 2007, 2009; Osterkamp and Hupp, 2010]. The spatial extent of vegetation cover within fluvial corridors has been shown to be strongly controlled by flow disturbance [Johnson, 2000; Lytle and Merritt, 2004; Camporeale and Ridolfi, 2006; Bertoldi et al., 2009, 2011]. Once the vegetation is established, root systems increase the erosion resistance and stability of fluvially deposited sediments [Pollen et al., 2004; Pollen-Bankhead et al., 2009; Tal et al., 2004; Eaton, 2006; Tal and Paola, 2007] and the flow resistance of plant canopies reduces flow velocities during inundation events, leading to the retention of sediment and organic matter [McKenney et al., 1995; Bennett et al., 2008]. The interac1 School of Geography, Queen Mary, University of London, London, UK. 2 Department of Geography, King’s College London, London, UK.

Copyright 2011 by the American Geophysical Union. 0043-1397/11/2010WR010319

tion between these disturbing and resisting factors leads to the development and turnover of a range of important fluvial landforms including floodplains, islands, scroll bars, and benches. To date, laboratory experiments [e.g., Gran and Paola, 2001; Coulthard, 2005; Tal et al., 2004; Braudrick et al., 2009; Tal and Paola, 2010] and developments in numerical modeling [e.g., Millar, 2000, 2005; Murray and Paola, 2003; Camporeale and Ridolfi, 2006; Perucca et al., 2006, 2007; Perona et al., 2009] have provided quantitative and spatially distributed evidence for the importance of these two-way interactions for river channel form and dynamics. However, quantification of the morphological effects of vegetation colonization and growth in the field has been based mainly upon ground measurements drawn from discrete spatial units such as transects across a river’s active corridor; regular, rectangular quadrats; or individual landforms such as floodplain edges, islands, and bars [e.g., Gurnell et al., 2001; Kalischuk et al., 2001; Samuelson and Rood, 2004; Corenblit et al., 2009]. [3] In this paper, we provide a detailed, continuous, spatial analysis of vegetation properties, landforms, and associations between vegetation and landforms along a 21 km reach of the braided Tagliamento River in northeast Italy. This is achieved by combining information from airborne

W06525

1 of 13

W06525

BERTOLDI ET AL.: TOPOGRAPHIC SIGNATURE OF VEGETATION

lidar data, digital color air photographs, and ground observations. Lidar data (often complemented by other remotelysensed data sets) were recently used to separate the vegetation canopy from the underlying topography; to classify forests and other vegetation ground cover and habitats ; to estimate vegetation roughness; and to identify forest successional stages [e.g., Antonarakis et al., 2008; Straatsma and Baptist, 2008; Falkowski et al., 2009; Geerling et al., 2009]. These data have allowed us to explore the degree to which the morphology of this braided river changes as vegetation extent and development also change. Specifically, we demonstrate that the shape of the elevation frequency distribution of the active corridor changes from being negatively skewed and peaked (leptokurtic) to being symmetrical and platykurtic, but also showing secondary peaks as vegetation extent and height (maturity) increase. We interpret this change in form as being primarily driven by aggradation of sediment around the developing vegetation to form distinct topographic islands; but this then has the effect of concentrating flow energy into a narrower channel width at intermediate to high flow stages [Tal et al., 2004; Tal and Paola, 2007], leading to a potential for change in the form of the unvegetated areas.

2.

Study Site

[4] The Tagliamento is a large, gravel bed, braided river located in the Friuli Venezia Giulia region, in northeast Italy. It drains from the Alps to the Adriatic Sea, spanning alpine to Mediterranean climatic regimes, and remaining essentially morphologically intact through the majority of its course. The present study focuses on a 21 km reach extending between river kilometers 67 and 89 (Figure 1), for which an airborne lidar survey and digital color air pho-

W06525

tograph cover were available following a survey in May 2005. Within the study reach, the braided course of the river extends to a maximum width of 1 km and also includes a narrow gorge at river kilometer 83, which forces the corridor to shrink to a minimum width of 130 m. This natural constriction causes significant changes in the river’s morphological, hydrological, and hydraulic behavior. [5] Patches of riparian shrubs and trees of widely varying size and age are found throughout the reach [Zanoni et al., 2008]. Populus nigra (black poplar) is the dominant riparian tree species, although several willow species (Salix alba, S. daphnoides, S. eleagnos, S. purpurea, and S. triandra) are also common, particularly S. eleagnos [Karrenberg et al., 2003]. Vegetative regeneration, particularly from uprooted trees deposited on gravel bars during floods, is a key process in the development of vegetated patches along the river’s active corridor [Gurnell et al., 2001, 2005; Gurnell and Petts, 2006], leading to the development of wooded islands covered by continuous, dense, mature trees and surrounded by open gravel or water. As a result of the varied hydraulic (disturbance) and hydrological (groundwater) conditions along the reach, vegetation cover is highly variable in both space and time, typically ranging from near 0% up to 40% [Bertoldi et al., 2011]. Therefore, the reach provides an ideal study area in which to evaluate the effect of vegetation on river morphology, particularly since other hydraulic parameters (flood discharge, slope, grain size) remain almost constant.

3.

Methods

[6] Remotely sensed data are increasingly available, and provide a potentially valuable source of information for both river scientists and managers. In the present analysis,

Figure 1. Location of the study reach and the 19 analyzed subreaches. Inset depicts a map of Italy and location of the study site. 2 of 13

W06525


we assess the possibility of integrating different data types and techniques to provide a reliable methodology for analyzing morphological processes in large, shallow rivers to meet both scientific and operational needs. Our analytical approach is summarized in Figure 2 and involves coupling the analysis of airborne lidar and digital true color air photograph data captured in May 2005 by the U. K. Natural Environment Research Council (NERC), with ground survey data captured between 2007 and 2010, to obtain a highresolution description of river morphology, with quantification of both bed elevation and vegetation characteristics. The analysis employed the free software FUSION, developed by the U. S. Department of Agriculture, Forest Service, Remote Sensing Applications Center (available at http://www.fs.fed.us/eng/rsac/) to analyze raw lidar data. This supported the extraction of a digital elevation model (DEM), the average tree height and vegetation density, allowing estimation of the proportion of the active corridor covered by trees and shrubs (see section 3.1). The water depth was estimated from aerial photographs (with ArcMAP software) to complete the DEM extracted from the lidar data (see section 3.2). Moreover, several field campaigns carried out between 2007 and 2010 provided data that helped to parameterize and validate these methods (section 3.3). In particular, we used ground-surveyed cross profiles of the active corridor and of specific landforms, and we also measured tree heights to check the accuracy of the DEM and the vegetation filtering process. The lack of synchronicity in the capture of ground and airborne data is a common problem in many studies that use archive rather than purpose-collected surveys [e.g., Lane et al., 2010]. The combined analysis of remotely sensed and ground data from different dates demanded some rather unconventional analytical approaches, but the ground data proved crucial for testing and calibrating metrics estimated from the airborne surveys. [7] The study site was then divided into 21 one-km long subreaches in order to investigate the differences in the

W06525

river bed topography under a range of different vegetation conditions. We computed the frequency distribution of bed elevation within the entire active corridor and then compared parameters of this frequency distribution with parameters of vegetation extent, distribution, and development. 3.1. Lidar Data Analysis [8] The lidar data and color air photographs were acquired during May 2005 providing full coverage of the 21 km study reach with a density of approximately 1.05 points m2. The FUSION software employed in the lidar data analysis implements the widely tested hierarchic robust filtering technique described in Krauss and Pfeifer [1998] to filter the data and construct the DEM. This method provides an automatic classification of the lidar points into terrain and vegetation, where points are more likely to be classified as vegetation if they protrude from the neighboring point cloud. The procedure has been found to work quite well, particularly when the ground is not characterized by steep slopes and the vegetation is sparse enough for at least 25% of the points to reach the ground [Hollaus et al., 2006], conditions that are generally met in the study area, particularly at the time of survey during late spring, when the foliage is not fully developed. As the main characteristics of terrain and vegetation composition are similar, this iterative method was applied on a 4 m 4 m grid with the same input parameters across the entire 21 km reach to produce a DEM on a 2 m grid. [9] The longitudinal slope was then estimated by computing a moving average of the bed elevation based on analyses within an 800 m square window (800 m approximates the typical width of the active river corridor). Within each 800 m square window, areas outside of the active corridor were excluded when calculating the average bed elevation. The slope was then subtracted from the DEM to allow direct comparison of the bed topography between different subreaches. Figure 3a shows a small area of the DEM, illustrating the occurrence of bars and islands, as

Figure 2. Flowchart showing the stages in the analysis of the lidar data, digital color images, and ground data. 3 of 13

W06525


W06525

Figure 3. (a) A small extract from DEM. (b) The estimates of vegetation height for the same area shown in Figure 3a. (c) Box and whisker plots of differences between the elevation estimated from the lidar DEM and the ground surveys along three transects located in Figures 3a and 3b. (d) Comparison of the ground measured and lidar-computed tree heights within 30 quadrats distributed along the study reach. well as many fine-scale features on these landforms, such as small channels, patches of sediment, and avalanche face deposits. [10] The vegetation height was computed as the difference between the interpolated ground surface (DEM) and the lidar point cloud in a 5 m 5 m regular grid (e.g., Figure 3b). The vegetation density was computed as the proportion of points within the 5 m grid that lay at least 1 m above the ground surface. This allowed for an estimation of the proportion of the active corridor that was vegetated. The density of vegetation taller than 5, 10, and 20 m was computed in a similar way to characterize the vegetation profile. 3.2. Water Depth Estimation [11] In order to fully analyze associations between river morphology and riparian vegetation, it was necessary to characterize the topography of the entire river corridor, including any areas submerged by water. Gravel bed braided rivers are characterized at low flow by the presence of quite small, inundated areas, where it is not possible to obtain bed elevation data with a standard lidar survey because the dominant reflection is from the water surface. About 13% of the area of the river’s active corridor was covered by water at the time of the 2005 survey, which under such base flow conditions generally does not exceed 1 m in depth. To estimate water depth we used an optical remote sensing technique suitable for retrieving shallow depth information [e.g., Winterbottom and Gilvear, 1997; Marcus and Fonstad, 2008; Leigleiter and Roberts, 2009; Leigleiter et al., 2009]. A ratio-based method was employed in order to detect changes in depth and filter out the effect of changes in bottom albedo [e.g., Dierssen et al., 2003; Mishra et al., 2007]. Legleiter et al. [2004] and Marcus and Fonstad [2008] demonstrated that the following log-transform of the red-over-green band

ratio correlates linearly with water depth across a wide range of substrate types: D ¼ a ln

1 ; 2

ð1Þ

where D is water depth, a is a constant, and 1 and 2 are the intensity of the red and green bands. [12] Figure 4 illustrates the application of this method to a small area of the study reach. Figure 4a shows the true color air photo for an area where water of different depth is present (the main channel at the bottom of the frame and a few secondary branches). Figure 4b displays the logarithm of the ratio between the green and red bands. The presence of vegetation can disturb the water depth computation, as in some cases small patches of vegetation have a similar ratio value to the water. We used the vegetation cover maps derived from the lidar data to mask out areas covered by vegetation (Figure 4c), allowing us to extract only the inundated areas (Figure 4d). Finally, we needed to estimate the coefficient a in equation (1) in order to compute the water depth. This step is usually based on ground measurements of water depths at the time of the survey. Because such data were not available, we calibrated the a coefficient by referring to the topography of four cross sections surveyed 2 years later [see Bertoldi et al., 2010]. The 2005 lidar cross sections showed different topography in some parts of the profile when compared to 2007 field survey data because of intervening fluvial erosion and deposition (e.g., Figures 5a and 5b). Therefore, we estimated the linear coefficient a in equation (1) by minimizing the difference between the standard deviation of the elevation of a set of equally spaced points extracted from the ground surveyed

4 of 13

W06525


Figure 4. Example of the water depth estimation procedure. (a) Digital image ; (b) logarithm of the red-over-green ratio (see equation (1)); (c) application of the vegetation mask obtained from the lidar data; (d) estimated water depth map.

Figure 5. Calibration of the water depth estimates. (a and b) Two examples of cross sections showing the 2007 ground survey in comparison with the 2005 profile extracted from the DEM. (c) Comparison of the cross-sectional standard deviation of elevation derived from the ground survey with the standard deviation computed from the DEM with (black dots) and without (gray triangles) the addition of the estimated water depth. 5 of 13

W06525

W06525


cross profiles and from the lidar data corrected by the water depth estimation procedure (e.g., Figure 5c). Cross profiles extracted from the DEM without the water depth procedure (which included flat water surfaces), were characterized by a much lower standard deviation (gray triangles in Figure 5c) than those with the subwater surface profile added (black dots in Figure 5c), and the water depth correction provided cross profiles which have visually similar characteristics to the ground surveys, although there has clearly been channel movement in the intervening period between collection of the lidar and field topographic survey data (e.g., Figures 5a and 5b). If possible, this procedure should be repeated for each of the 15 air photos covering the 21 km long reach. However, water depth data were only available for one of the photographs, therefore we used this photograph to estimate the coefficient a for the other photographs. [13] To test the sensitivity of the entire active corridor DEM (i.e., the exposed and inundated area) to extreme differences in the linear coefficient a in equation (1), we estimated some error margins for two parameters of the active corridor elevation frequency distribution: skewness and kurtosis. First, we changed the value of the a coefficient in the water depth computation by 610% in each subreach. On average, the skewness of the elevation frequency analysis changes by 60.05%, whereas kurtosis changes by 60.1%. Second, we modeled a random error in each cell, changing the water depth in a range 625%. This resulted in changes in skewness by up to 60.01% and in kurtosis by up to 60.04%. These two approaches take into account possible local errors in the computation (up to 25%) and bulk over/underestimates in each subreach by 10%. 3.3. Analysis of Vegetation and Topographic Field Data [14] The accuracy of the obtained DEM for heavily vegetated areas was checked through comparison with 2010 ground surveys of three cross profiles of an island that had not been affected by river inundation since the 2005 survey date (Figure 3b). The comparison was undertaken in the most unfavorable area, where the presence of steep banks and mature vegetation cover may violate the assumptions behind the filtering procedure. Box and whisker plots summarize the results of the comparison (Figure 3c), illustrating that half of the surveyed points (box) show an error lower than 30 cm, with only 10% of the points showing an error larger than 40 cm. Thus, estimating surface topography in vegetated areas proved to be quite accurate in most cases, because the density of points reaching the ground was sufficiently high. However, steep banks (with a nearvertical step in bed elevation of up to 2 m) were smoothed by the filtering procedure, leading to the largest errors of up to 40 cm. Bare gravel points (that represent up to 90% of the total active river corridor area) showed an error lower than 15 cm. [15] The accuracy of the lidar estimates of tree height were evaluated by comparison to field measurements acquired in 2010. The tree height was sampled with a clinometer in 30 quadrats of different vegetation height and density distributed along the reach and geographically located using a GPS. Comparison between the tree heights estimated from the lidar data and from ground measurements (Figure 3d) show that the field-surveyed trees are

W06525

consistently taller, with a difference of up to 5 m. This is because of tree growth between 2005 and 2010. The ground data on tree growth rates in 2010, on the basis of tree ring counts at 1 m above ground level of samples of 20 three-m tall P. nigra individuals at nine sites along the reach, were linearly interpolated along the reach to provide estimates that would allow a standardized measure of tree growth to be incorporated into some of the analyses. The sampled trees were predominantly single-stemmed, relatively isolated specimens that were unlikely to be affected by competition from surrounding trees. At seven of the sampled sites the trees were located on the tops of the highest bars to ensure a comparable topographic position between sites, although at two sites, where sufficiently tall trees were not available within the active corridor, trees were sampled along the margin of the floodplain. Although the origin of the trees was unknown (seedlings or vegetative propagules), sampling of tree rings at 1 m above ground level ensured that growth was estimated from a baseline at which the tree’s roots would have been well established (particularly if the initial ground surface had been buried by subsequent sediment deposition), and thus the estimated growth rates are unlikely to be significantly affected by propagule type. Overall, the aim of selecting a substantial sample of isolated trees of the same species, height, and topographic position, and then estimating their growth rate from 1 m above the ground surface was to ensure that sample site comparability in the estimated growth rates would be indicative of the relative growing conditions along the reach. Analyses of these data were largely based on graphical visualizations coupled with the estimation of nonparametric (Spearman’s rank) correlations between growth rate and other properties to test the strength of observed graphical relationships. Results suggest growth rates can be as much as 1 m yr1 and typically fall in the range of 20 to 60 cm yr1, thus explaining the differences between measured tree heights and those estimated by the lidar 5 years earlier (Figure 3d).

4.

Results

[16] Having extracted information on vegetation cover and height, and also the morphology of the river’s active corridor along the entire study reach, analysis of these data focused on 1 km long subreaches (Figure 1) in order to investigate spatial variations in vegetation and morphological properties. Since subreaches 9 and 15 were not entirely alluvial (subreach 9 contains a large outcrop of bedrock, subreach 15 includes a narrow bedrock gorge) they were excluded from these analyses, leaving 19 subreaches where interactions between vegetation and morphology could be investigated. 4.1. Properties of the Vegetation [17] We focused on the vegetated areas within each subreach and computed the average canopy height and vegetation density estimated from the lidar data for each subreach. This analysis revealed large contrasts in vegetation properties between subreaches with no vegetation cover over 5 m tall in subreaches 2 and 3 (Figure 6a) and large variations in average canopy height and density (Figure 6b). There was a strong positive relationship (Spearman’s rank

6 of 13

W06525


W06525

Figure 6. Vegetation metrics computed from the analysis of lidar data. (a) Variations in proportion of the active corridor occupied by vegetation taller than 5 m between the 19 analyzed subreaches of the 21 km study reach. Subreach 9 was excluded because of extensive bed rock exposure and subreach 15 did not support any detectable vegetation cover within the active corridor. (b) Variations in reach-averaged canopy height and vegetation density. Relationship between reach-averaged canopy height and (c) the proportion of the active corridor occupied by vegetation taller than 5 m. (d) Reach-averaged vegetation density. correlation ¼ 0.812, p < 0.001) between vegetation height and cover of tall (>5 m), vegetated patches (Figure 6c), and subreaches containing taller vegetation also displayed a higher average vegetation density (Figure 6d, Spearman’s rank correlation ¼ 0.753, p < 0.001). All of these relationships illustrate that the study reach provides large contrasts in tree cover, height, and density within the active corridor between subreaches that provide a good gradient of vegetation properties against which changes in morphology can be explored. 4.2. Comparisons Between Vegetation and Morphological Properties of Subreaches [18] Because valley slope was filtered from the DEM and elevations were expressed as deviations from the local average elevation within the active corridor, elevationfrequency relationships could be directly compared between subreaches. At the same time, the detailed spatial data on vegetation properties allowed the elevation frequency of the ground surface and overlying vegetation of varying height to be superimposed on the same diagram. Figure 7 compares elevation frequency diagrams of the active corridor surface and of the distribution of vegetation of different heights associated with those surface elevations in two subreaches with strongly contrasting vegetation extent. The frequency distributions of bed elevation are very different

in the two reaches. The reach with a lower vegetation extent (Figures 7a and 7c) displays a more compact, peaked elevation distribution than the more heavily vegetated reach (Figures 7b and 7d). In both reaches, vegetated patches start to appear at an elevation just below the average and become very noticeable at elevations higher than 0.5 m above the average, with bare surfaces virtually disappearing above an elevation of 1 m above the average. The more heavily vegetated subreach also displays more mature trees than the less vegetated subreach, with trees taller than 10 m found at elevations over 1 m above average elevation. [19] In order to characterize the frequency distribution of bed elevation with a single parameter we followed the procedure used by Paola [1996] and Nicholas [2000] to describe the transversal variability of shear stress (or bed elevation) in gravel bed braided rivers by fitting a gamma distribution: pð H Þ ¼

H 1 eH ; ðÞ

ð2Þ

where H is the dimensionless bed elevation (expressed as elevation above the lowest point and standardized with the mean value), is the Euler gamma probability density function, and is the parameter describing the shape of the distribution. We computed the best fit value of the parameter for each of the subreaches by minimizing the squared error

7 of 13

W06525


W06525

Figure 7. Frequency distribution of the bed elevation for two subreaches. (a and b) The ground surface elevation frequency distribution of the active corridor, with each bar subdivided according to the proportion of pixels at that elevation occupied by bare gravel or vegetation of one of four height classes. (c and d) Separate elevation frequency distributions of pixels occupied by gravel or one of the four vegetation height classes.

between the observed elevation data and that estimated from the distribution of p(H). The best fit estimates of ranged between 2 and 9 and showed a strong relationship with the standard deviation of bed elevation, suggesting a generally good characterization of the variability in bed topography (Figure 8), although the degree of correspondence to the gamma distribution varied between subreaches.

Figure 8. The relationship between the parameter of the fitted gamma distribution and the bed SD for the 19 subreaches.

[20] Four subreach comparisons between the frequency distribution of observed bed elevation and the best fit gamma distribution are presented in Figure 9. Figures 9a and 9c represent two subreaches (3 and 8) where vegetation cover is very low (1.2% and 1.8%, respectively). Here the gamma function provides a good characterization of the bed topography. In contrast, Figures 9b and 9d represent two subreaches (5 and 12) with more extensive vegetation cover (8% and 11%, respectively). These subreaches are not well represented by the gamma function. Lower values of were needed to describe these broader distributions of elevation, but the distributions also show secondary peaks. These secondary peaks are enhanced by the logarithmic scale of the plots in Figure 9, but are also discernible, although very subdued, in the elevation frequency distributions which are not plotted on a logarithmic scale (e.g., Figure 7b). Because of these secondary peaks, the reaches with more extensive vegetation cover cannot be well characterized by the single-peaked gamma function. In Figures 9b and 9d, the secondary peaks occur at relatively high elevations and correspond with areas of well-developed vegetation, indicating a clear topographic signature related to vegetation development. This interpretation is further supported by the significant positive relationship between the presence of relatively high areas within the active corridor and the proportion of the active corridor occupied by

8 of 13

W06525


W06525

Figure 9. The dimensionless bed elevation frequency distribution and best-fit gamma distribution for four example subreaches supporting different amounts of vegetation. mature (> 5 m tall) vegetation (Figure 10, Spearman’s rank correlation ¼ 0.559, p ¼ 0.011). [21] Further analysis of associations between the bed elevation frequency distribution and vegetation properties was undertaken to characterize the degree to which different

Figure 10. The relationship between the proportion of the active corridor occupied by vegetation taller than 5 m and the proportion with a bed elevation higher than 1 m above the average level.

vegetation properties appeared to affect the braided morphology across the 19 subreaches. Three properties of the untransformed bed morphology frequency distribution (standard deviation, skewness, and kurtosis) were investigated. Four scatter plots were constructed for each of these morphological properties, to explore associations with four properties of the vegetation : (1) interpolated average 3 m tall P. nigra growth rate (observed in 2010); (2) the proportion of the active corridor occupied by relatively tall (>5 m) vegetation ; (3) the median elevation of vegetated pixels; and (4) the average canopy height. Associations between the standard deviation of bed elevation and the four vegetation properties were weak and two were not statistically significant (Table 1), so the scatter plots are not presented. However, all the correlations between the skewness and kurtosis of the bed elevation frequency distribution and the vegetation variables were statistically significant (Figures 11 and 12; Table 1), with correlations in excess of 0.7 for several vegetation indices, particularly when correlated with skewness. Moreover, the error bars shown around each point on the scatterplot show that these associations are not adversely affected by the inclusion of a range of different values for the a coefficient (equation (1)) when estimating the inundated component of the DEM. These statistically significant relationships demonstrate that as vegetation growth rate, extent, median elevation and height increase, the distribution in bed elevation changes from being negatively skewed in reaches with little vegetation cover where tree growth rates are low to showing negligible skew in more heavily vegetated reaches where tree

9 of 13

W06525

W06525


Table 1. Spearman’s Rank Correlation Coefficients (p-values) Between Four Vegetation Metrics and Three Properties of the Bed Elevation Frequency Distribution Across the 19 Subreachesa SD (m) 1

Tree growth rate (cm yr ) Proportion with vegetation >5 m Median elevation of vegetated pixels (m) Average canopy height (m) a

Skewness b

0.37 (0.066) 0.40b (0.046) 0.45 b (0.025) 0.16 (0.236)

0.68 (0.002) 0.53b (0.01) 0.82b (