1 Mapping the spatio-temporal distribution of key vegetation cover properties in lowland river 1 reaches, using digital photography 2 Authors: Veerle Verschoren1*, Jonas Schoelynck1, Kerst Buis1, Fleur Visser2, Patrick Meire1, Stijn 3 Temmerman1 4 5 1. University of Antwerp, Department of Biology, Ecosystem Management Research Group, 6 Universiteitsplein 1C, B-2610 Wilrijk, Belgium. 7 2. University of Worcester, Institute of Science and the Environment, Henwick Grove, 8 Worcester WR2 6AJ, UK 9 10 *Corresponding author 11 UA - Campus Drie Eiken 12 Ecosystem Management research group 13 Universiteitsplein 1 14 Building C, C1.20 15 B - 2610 Wilrijk, Belgium 16 17 verschoren.veerle@hotmail.com 18 Tel +32 3 265 22 52 19 Fax +32 3 265 22 71 20 21 2 Abstract 22 The presence of vegetation in stream ecosystems is highly dynamic in both space and time. A 23 digital photography technique is developed to map aquatic vegetation cover at species level, which 24 has a very-high spatial and a flexible temporal resolution. A digital single-lens-reflex (DSLR) 25 camera mounted on a handheld telescopic pole is used. The low-altitude (5 m) orthogonal aerial 26 images have a low spectral resolution (Red-Green-Blue), high spatial resolution (~1.9 pixels cm-2, 27 ~1.3 cm length) and flexible temporal resolution (monthly). The method is successfully applied in 28 two lowland rivers to quantify four key properties of vegetated rivers: vegetation cover, patch size 29 distribution, biomass and hydraulic resistance. The main advantages are that the method is: (i) 30 suitable for continuous and discontinuous vegetation covers (ii) of very-high spatial and flexible 31 temporal resolution, (iii) relatively fast compared to conventional ground survey methods, (iv) 32 non-destructive, (v) relatively cheap and easy to use, and (vi) the software is widely available and 33 similar open source alternatives exist. The study area should be less than 10 m wide and the 34 prevailing light conditions and water turbidity levels should be sufficient to look into the water. 35 Further improvements of the images processing are expected in the automatic delineation and 36 classification of the vegetation patches. 37 38 Key words: macrophytes, vegetation cover, very high spatial resolution, flexible temporal 39 resolution 40 41 3 Introduction 42 The presence of aquatic vegetation in river ecosystems tends to be highly variable in space and 43 time. Because of the importance of vegetation in fluvial ecosystems there is a need to efficiently 44 map and monitor this variability. The study described in this paper presents a method for detailed 45 mapping of the dynamic vegetation patterns in rivers. 46 47 Macrophytes, or aquatic plants, have different growth forms: exclusively submerged, submerged 48 with floating leaves, exclusively floating or emergent. They occur in single species beds with a 49 continuous cover or in a discontinuous composition of multiple species. The interaction between 50 vegetation and water flow leads to spatial patterns of vegetation patches at reach scale, river 51 sections of 100 to 200 m (Schoelynck et al. 2012). A macrophyte patch can be defined by an area 52 covered by vegetation, which has a finite spatial extent that is larger than an individual shoot but 53 smaller than the entire reach. The size of these vegetation patches varies strongly from a few square 54 decimetre to a few square meter (Gurnell et al. 2006; Sand-Jensen et al. 1999). The size of the 55 individual leaves ranges from several square centimetre to several square decimetre. In temperate 56 mid-latitude climate zones, the development of these vegetation patches has an annual cycle with 57 abundant plant growth in the growth season followed by die-back (Battle and Mihuc 2000; 58 Menendez et al. 2003). 59 60 These dynamic growth processes result in frequent changes in key properties of vegetated rivers 61 including vegetation cover, patch size distribution, biomass and hydraulic resistance. These 62 properties in turn affect stream processes, such as: nutrient cycling (Dhote and Dixit 2009; Krause 63 et al. 2011; Seitzinger et al. 2006), the transport of dissolved matter and the retention of particulate 64 4 matter (Cordova et al. 2008; Horvath 2004; Lamberti et al. 1989), bedload sediment transport 65 (Gibbins et al. 2007) and drift of macro-invertebrates (Extence et al. 1999). 66 67 The first of the key properties, macrophyte cover, is an essential parameter used for monitoring of 68 fluvial ecosystems. Macrophytes are for example used as a quality parameter in the assessment of 69 the ecological status of surface water for the Water Framework Directive in Europe (EU 2000). 70 This assessment takes into account the number of species and species abundance. The second key 71 property, the frequency distribution of patch sizes, can be used to investigate spatial self-72 organisation in river ecosystems. Spatial self-organisation in rivers is the process where large scale 73 patterns develop from disordered initial conditions through small scale feedbacks between plants 74 and the water flow (Lejeune et al. 2004; Rietkerk et al. 2004; Schoelynck et al. 2012). The process 75 is important for ecosystem functioning, since self-organised ecosystems have a higher resilience 76 and resistance to environmental change and a higher productivity compared to homogeneous 77 ecosystems (van de Koppel et al. 2008). Schoelynck et al. (2012) showed the presence of spatial 78 self-organisation of macrophytes patches in lowland rivers. They demonstrated that the size 79 distribution of macrophytes patches can be described by a power-law relationship, which is an 80 indication of self-organisation (Newman 2005; Scanlon et al. 2007). Thirdly, biomass is a crucial 81 parameter in many ecological studies for example for the calculation of mass balances or 82 quantification of nutrient fluxes (Borin and Salvato 2012; Dinka et al. 2004). The parameter values 83 will depend on vegetation extent and species composition. Finally, the hydraulic resistance of a 84 river reach is influenced by obstructions like aquatic vegetation, bed material, the meandering of 85 the river and irregularities in its cross-sections (Chow 1959). Macrophytes increase the hydraulic 86 resistance which leads to reduced stream velocities and increased water levels upstream (De 87 Doncker et al. 2009b). A direct effect of increased water levels is a higher risk of flooding. The 88 5 effect of macrophytes on the hydraulic resistance is threefold: through vegetation density (e.g. 89 biomass (De Doncker et al. 2009b)), plant characteristics (e.g. growth form (Bal et al. 2011)) and 90 spatial distribution (e.g. cross-sectional blockage (Green 2005b)). In general: high biomass, stiff 91 plants and large cross-sectional blockage all lead to a higher resistance to water flow, which is 92 expressed by a higher Manning roughness coefficient (n) (Chow 1959; Madsen et al. 2001; 93 Vereecken et al. 2006). Recently more detailed hydrodynamic models have been developed, which 94 incorporate such plant features (Verschoren et al. 2016). 95 96 To quantify the above-mentioned vegetation parameters and use them for monitoring, modeling 97 and management of river processes, a method is needed that can efficiently map the dynamic 98 patchiness of macrophytes in rivers with a very-high spatial (subcentimetre) and flexible temporal 99 resolution. The detection of fine scale details in structure, texture and pattern on very-high spatial 100 resolution image data allows identification of macrophytes up to species level (Bryson et al. 2013; 101 Visser et al. 2013). Properties like biomass and hydraulic resistance depend strongly on species 102 composition and need flexible temporal resolution (e.g. monthly) data acquisition to catch seasonal 103 variation. Low-altitude image data collection seems the most suitable method to obtain high spatial 104 and flexible temporal resolution data while minimizing the time and cost (Carter et al. 2005; 105 Legleiter 2003). 106 107 High resolution low-altitude image data collection techniques proved to be suitable for many 108 ecological studies in intertidal marine environments with a spatial extent between 0.01 - 1 ha and 109 resolutions ranging between 0.5 - 5 cm. Examples are patterns of algae distribution (Guichard 110 2000), biophysical control of benthic diatom films and macroalgae (van den Wal 2014), the 111 distribution of eelgrass and blue mussel (Barrell and Grant 2015), and terrain models of intertidal 112 6 rocky shores (Bryson 2013). However, images were mostly obtained at low tides while study sites 113 were not inundated. Due to the absorption of light in water (Visser et al. 2013), limited spatial 114 resolution or high costs (Flynn and Chapra 2014; Husson et al. 2014; Shuchman et al. 2013), it is 115 only relatively recently that more studies started looking at mapping aquatic vegetation in 116 submerged environments, including rivers and lakes (Anker et al. 2014; Silva et al. 2008; Villa et 117 al. 2015). Hyperspectral remote sensing is successfully used to measure the river morphology 118 (Tamminga et al. 2015), to map invasive aquatic vegetation in a delta (Hestir et al. 2008) and 119 submerged macrophytes and green algae in rivers (Anker et al. 2014). However, these 120 hyperspectral images are costly and/or have too low spatial resolution (~1-3 m) to be applied in 121 small streams (stream width <10 m) (Shuchman et al. 2013). 122 123 Recent efforts have been undertaken to obtain low-cost, high spatial resolution (subdecimetre to 124 submetre) images, but with a low spectral range. At a resolution of 25 cm, Flynn and Chapra (2014) 125 mapped aquatic submerged vegetation and green algae in small lowland rivers and lakes, and 126 Nezlin et al. (2007) mapped algae and mussels on tidal flats. Higher spatial resolution images were 127 obtained by Husson et al. (2014) (5.6 cm) and Anker et al. (2014) (4 cm) to record aquatic 128 vegetation. However, these resolutions are often still too coarse to distinguish different macrophyte 129 species, which sometimes requires assessment of the shape of individual leaves. A common 130 recommendation from several of the before mentioned studies is the requirement that images 131 should be taken under optimal conditions, e.g. no diffuse light, sun at its highest position, clear 132 water, no ripples. However, this almost never occurs in reality and therefore further limits the 133 applicability of the method and is an additional reason why this technique has not yet become 134 mainstream in river ecosystem research: it is difficult to look into a river trough a camera lens 135 (Visser et al. 2013). 136 7 137 In this paper we present a rapid and cost-effective digital aquatic vegetation cover photography 138 technique based on orthogonal low-altitude images with a very-high (subcentimetre) spatial 139 resolution and flexibility to collect data frequently (monthly or higher) under optimal weather and 140 scene illumination conditions (no diffuse light and the sun at its highest position). We use the 141 collected images to map the spatial distribution of aquatic vegetation at species level in two river 142 ecosystems (± 200 m river reaches) and we demonstrate how the maps are suitable to monitor four 143 key properties of vegetated lowland rivers, namely vegetation cover, patch size distribution, 144 biomass and hydraulic resistance. 145 Materials and methods 146 Study area 147 The data were collected in 2013 in two lowland rivers in the North East of Belgium: Zwarte Nete 148 and Desselse Nete (51° 15’ 3.45” N, 5° 4’ 54.27’’ E) (Fig. 1). Both rivers are characterised by 149 extensive plant growth in summer and are surrounded by pasture, which limits overhanging and 150 other riparian vegetation. The rivers have a low suspended matter concentration (< 50 mg L-1) and 151 the substrate consists of sand (median grain size of 167 µm). The Zwarte Nete has a mean width 152 of 4.4 m, water depth ranges between 0.5 - 0.6 m and discharge between 0.2 - 0.5 m3 s-1. A reach 153 of 187 m (821 m2) was mapped where multiple plant species were present. The Desselse Nete is 154 slightly larger with a mean width of 5.4 m, mean water depth of 0.6 - 0.7 m and mean discharge 155 between 0.3 - 0.6 m3 s-1. Here a reach of 180 m (1123 m2) was selected, dominated by a single 156 submerged species with floating leaves: Potamogeton natans (L.). The following species were 157 present in one or both reaches: submerged species: Callitriche obtusangula (Le Gall), 158 Myriophyllum spicatum (L.), Potamogeton pectinatus (L.) Ranunculus peltatus (L.), Sagittaria 159 8 sagittifolia (L.), Sparganium emersum (L.) and emergent species: Typha latifolia (L.) and riparian 160 vegetation (not identified to species level). No exclusively floating species were present. 161 162 Image collection 163 The images were collected with a Nikon D300s DLSR camera with a crop sensor 164 (NikonCorporation, 2009). As inherent to most unmodified cameras, images consisting of three 165 broad spectral bands are obtained (RGB): a blue (400 - 500 nm), a green (500 - 600 nm) and a red 166 band (600 - 700 nm). The files were compressed as JPEG (fine) with an image dimension of 4288 167 x 2848 pixels and an image size of 12.3 megapixels. The camera was equipped with a Tokina AT-168 X 116 Pro DX (11-16mm, F2.8) wide angle lens that has a large field-of-view and a distortion of 169 0.6% (Dxomark). The zoom was set to the widest possible angle and the focus at infinity. The 170 camera was attached with a ball head to a handheld telescopic pole to take low altitude images of 171 the water surface at nadir (Fig. 2a). The lower end of the pole was placed at the river bank. The 172 pole was tilted so that the camera was positioned above the center line of the river at a height of 173 approximately 5 m above the water surface (Fig. 2b). The camera was remotely operated from a 174 laptop (tethered capture), which also provided live view to ensure correct positioning of the image 175 footprint. Both river banks had to be visible in each image. No polarization filter was used as this 176 was not thought to have an effect with the camera at nadir positon. Camera ISO was set to 200 to 177 minimize the noise and a variable aperture to achieve a fast shutter speed (Pekin and Macfarlane 178 2009). The images generally covered an area of 10 m (along the stream) x 6.5 m (across) (Fig. 3a 179 and 3b). 180 181 Multiple images were collected at monthly intervals covering the entire reaches of both rivers from 182 April to September 2013. The distance between two consecutive images was 4 m to ensure 183 9 sufficient overlap (~30 % overlap). Data was collected on clear days around noon to achieve 184 optimal illumination conditions. The angle between the sun and the camera is approximately 40° 185 between 11 a.m. and 1 p.m. (summertime) in Belgium. Several ground controls points (GCPs) were 186 positioned along the reaches to allow georeferencing of the image mosaics. Both reaches are 187 bounded upstream and downstream by small bridges which were included as GCPs. Geographic 188 coordinates for the GCPs were obtained with a dGPS (Trimble R4 GNSS, Eersel, NL) with an 189 accuracy of 1 cm. The exact coordinates of the river banks were once measured with an electronic 190 theodolite (Total Station, Sokkia set 510k, Capelle a/d Ijssel, NL) with a spatial interval of 2 to 3 191 m. The coordinates of the river bank were considered as complementary GCPs, which are clearly 192 visible on the images. 193 194 Spatio-temporal vegetation cover 195 Three steps are needed to create vegetation maps at species level: (i) image dehazing and stitching 196 by month and reach, (ii) georeferencing of image mosaics, and (iii) manual delineation of 197 vegetation patches. 198 Firstly, haze was removed from the images with the Autopano Giga (v. 3.0, Kolor, Francin, FR) 199 software using the Neutralhazer Light Anti-Haze plug-in. The software was then used to create 200 image mosaics along the full river reaches, using image matching algorithms to match up 201 overlapping photographs. For around 10 % of the images the matching process seemed to be 202 affected by reflection, movement of the vegetation with the river current and a homogeneous 203 riparian margin. In these cases we manually added extra control points at matching locations in 204 both images. This hardly affected the time to stitch. The image mosaics were exported as a JPEG. 205 This protocol was repeated for the images of both reaches and for each month. Secondly, in ArcGIS 206 (v. 10.1, ESRI Inc, Redlands, USA) the image mosaics were georeferenced using a spline 207 10 transformation. It should be noted that the GCPs were not present in all images that formed a 208 mosaic. An example of georeferenced image mosaics is given in Fig 3c and 3d. Thirdly, polygons 209 were drawn manually delineating the vegetation patches. Advantages and limitations of this 210 approach are extensively discussed at the end of this paper. Patches consisted of a single species 211 and had a minimum size of 2 dm2. For each polygon the type of species was determined from the 212 image (Fig. 3e and 3f). The surface area of each polygon was calculated and summed to obtain the 213 total vegetation cover per reach and per species type. 214 215 The manual image classification was validated against independent field measurements of 216 vegetation presence. A conventional grid method (Anker et al. 2014; Champion and Tanner 2000) 217 was used to estimate macrophyte cover on the ground. A rectangular grid of 2.88 by 0.88 m (36 by 218 11 cells of 0.08 by 0.08 m) was placed at a fixed location monthly in both streams on the same 219 days the images were collected. The presence of macrophytes in each cell is recorded and 220 determined to species level. The image data was resampled to 0.08 m resolution with each cell 221 coded according to the dominant species. The overall accuracy is calculated by comparing the 222 species in each cell of both grids with a true or false evaluation. This is done per month per river. 223 The relative cover for each vegetation class is given for the months with a cover accuracy of less 224 than 95 %. 225 226 Patch size distribution 227 We tested if the frequency distribution of patch sizes can be approached by a power-law 228 relationship. We therefore used the inverse cumulative distribution which is the probability that a 229 patch size (S) is larger than or equal to s (Newman 2005; Scanlon et al. 2007): 230 𝑃 (𝑆 ≥ 𝑠) ~ 𝑠−𝛽 Eq. 1 231 11 with s the size of a patch and β the power-law exponent. A power-law relationship in this context 232 means that the sizes of patches varies strongly with many small patches and relatively few large 233 patches. R (R Core Team 2014) version 3.2.0, was used to fit a standard least squares regression 234 on the log-transformed data. 235 236 Biomass 237 A conversion factor between cover and biomass can be obtained from the literature (e.g. Flynn et 238 al. 2002; Madsen and Adams 1989). However the required input data weren’t available for species 239 in our study area, therefore the four dominant species in both rivers (C. obtusangula, M. spicatum, 240 P.natans, S. emersum) were sampled monthly to obtain the monthly conversion factor 241 biomass:cover (Tab. 7). Vegetation samples were collected at the date of image acquisition, 242 downstream from the studied reaches to not destruct the natural growth of the vegetation within 243 the study reaches. Each month, three replicates per species were sampled by manually removing 244 the above ground vegetation in a quadrant of 0.5 m x 0.5 m that was placed upon a monotopic 245 vegetation patch. The samples were oven dried (at 70° C for 48 h) and weighed afterwards (dry 246 weight, DW). It has to be noted that in May 2013, no sample could be taken for C. obtusangula. 247 Therefore the average was taken of values for April and June to estimate the biomass in that month. 248 The total cover per species per month was obtained through the image analysis. Then the biomass 249 (gDW) per species was calculated monthly by multiplying the species-specific conversion factor 250 biomass:cover (gDW m-2) with the corresponding cover (m2) . The biomass values were summed 251 for the whole reach and divided by the total surface of the reach to obtain the total biomass (gDW 252 m-2) averaged out over all species and over the whole river reach. Since three replicates were taken, 253 the total biomass consists of three values. 254 255 12 The applied image analysis method aims to quantify vegetation cover in a non-destructive way. 256 However, the validation of the total biomass required mowing of all the vegetation and is therefore 257 a destructive method. We only had the opportunity to use the mowing method in August. On 26 258 and 28 August 2013 the entire reach in the Desselse Nete and Zwarte Nete was mechanically 259 mowed by cutting most vegetation just above the sediment and removing it from the river. All 260 mowed vegetation from both reaches was immediately weighed (fresh weight, FW). A 261 representative subsample of the biomass consisting of a mixture of all species, was transported to 262 the lab. The subsample was weighed (FW), dried at 70°C for 48h and reweighed (DW). This 263 enabled us to determine a conversion factor between FW and DW for the biomass of the entire 264 reach. R 3.2.0 was used to perform a one-sample t-test to test the difference between the total 265 biomass obtained by the mowing method (one value) and by the image method (mean with standard 266 error based on three values). 267 268 Hydraulic resistance 269 The hydraulic resistance of rivers can be expressed as a Manning coefficient (Chow 1959). The 270 commonly used equation to calculate the Manning coefficient is based on hydraulic parameters and 271 is applicable in vegetated and non-vegetated rivers (Eq. 2, Tab. 1) (Chow 1959). The equation uses 272 the cross-sectional area, discharge, hydraulic radius and the water level slope. The water discharge 273 was measured upstream of both reaches at the same days the images were taken, using an 274 electromagnetic flow meter (Valeport model 801, Totnes, UK) and calculated by the velocity-area 275 method (Bal and Meire 2009). Simultaneously, the water level was measured with two pressure 276 sensors (Eickelkamp, Geisbeek, NL) placed in the water column near the bridges bordering the reach 277 upstream and downstream with a time interval of 20 min. and with an accuracy of 0.5 cm. The 278 elevation difference between the pressure sensors was measured with a RTK-GSP. The water levels 279 13 are corrected for atmospheric pressure and averaged over 24 h for each sampling campaign. The 280 water level slope in the reaches was calculated by subtracting the upstream and downstream water 281 level, divided by the length of the reach. Additionally, different empirical relationships are used to 282 convert vegetation properties to the Manning coefficient. Based on the data of the surface area 283 coverage of Green (2005a) we found an empirical relationship (Eq. 3, Tab. 1) between the Manning 284 coefficient and the vegetation cover. De Doncker et al. (2009a) fitted an equation (Eq. 4, Tab. 1) 285 based on measurements of the biomass (gDW m-2) and the Manning coefficient. These empirical 286 relationships (Eq. 3 and Eq. 4) are easy to use, but have a limited application potential. They don’t 287 account for the species composition and the horizontal and vertical distribution of the vegetation and 288 are derived for a specific study area. The general Manning coefficient (Eq. 2) is used to validate the 289 empirical equations (Eq. 3 and Eq. 4). 290 Results 291 Between 86 and 115 images (~ 1.9 pixels cm-2, ~ 1.3 cm edge length) were taken per reach from 292 which 41 to 56 were selected to construct the image mosaic. The images collection took around 293 one hour per reach per sampling campaign. Reduced illumination of the submerged vegetation 294 target for the April and September data due to low sun angles, made macrophytes less visible in 295 the images. Delineation of the vegetation patches was still possible but the vegetation cover may 296 have been underestimated. Processing of the images took around two days for months with a low 297 vegetation abundance (< 30%) and around three days for months with a high vegetation cover (> 298 30%). 299 300 14 Spatio-temporal vegetation cover 301 The total vegetation cover and partial species cover is given per month for the two reaches (Fig. 302 4). In the Zwarte Nete, the total vegetation cover increases from April to August and suddenly 303 decreases in September due to the scheduled mowing event on 28 August 2013. The dominant 304 species in the Zware Nete are S. emersum and M. spicatum during the sampling period (Fig. 4a). 305 The natural development of the vegetation cover in the Desselse Nete is different. The growth was 306 disturbed by an extra mowing activity on 25 June 2013 for management and safety regulations. 307 Two months later, a scheduled mowing event took place on 26 August 2013. P. natans is the most 308 abundant species in the Desselse Nete each month and recovered completely 8 weeks after the first 309 mowing event (Fig. 4b). 310 The validation of the image method with the ground survey showed that the accuracy of species 311 identification is very high (>97 %) in the study reach dominated by a single species (Desselse Nete) 312 (Tab. 2). These high values are due the relative simple composition of the vegetation patches, where 313 the whole reach is covered by a single species. On the contrary, the accuracy is less (> 59%) in the 314 river with a heterogeneous composition of multiple species, certainly in months when the 315 vegetation patches are developing (June and July). So the accuracy to determine the exact location 316 of vegetation patches is limited in those months. 317 For those months with a cover accuracy less than 95 %, the relative cover of each vegetation class 318 is given separately in Tab. 3. The difference in cover between the ground survey method and the 319 image method for each vegetation class is less than 12 %. This means that the cover per vegetation 320 class agrees well between both methods. 321 322 15 Patch size distribution 323 In total 262 vegetation patches were mapped in August in the Zwarte Nete, of which 143 were C. 324 obtusangula patches. The surface area of these patches ranged between 0.04 m2 and 2.76 m2. The 325 size frequency distribution of the patches is plotted on a double logarithmic scale (Fig. 5). A 326 significant power-law relationship was found for the upper part of the distribution (least squares 327 regression on the log-transformed data; p< 0.001, R2 = 0.99; 59 % of the data). 328 329 Biomass 330 The total biomass per reach is estimated with the image analysis method on a monthly basis (Tab. 331 4). The monthly conversion factors are given in Tab. 7. The mowed vegetation is immediately 332 weighed (FW) and converted to dry weight with measured the conversion factor FW:DW equal to 333 10.3. The total biomass (gDW m-2) obtained by the image analysis method does not significantly 334 differ from the biomass (gDW m-2) obtained by the mowing method. The results of the one-sample 335 t-test is a p-value of 0.797 and 0.198 for the Zwarte Nete and Desselse Nete, respectively. 336 337 Hydraulic resistance of vegetated rivers 338 Variation of the Manning coefficient over time is shown for the Zwarte Nete and Desselse Nete in 339 Figure. 6. In the Zwarte Nete the Manning coefficient is based on hydraulic data, Eq. 2, increasing 340 from April to August and decreasing in September to values similar to those of April. The Manning 341 coefficients of the Zwarte Nete calculated with the empirical equations (Eq. 3 and Eq. 4) are well 342 in agreement. The largest difference is found in August with values of 0.26, 0.30 and 0.20 for Eq. 343 2, Eq. 3 and Eq. 4, respectively. The Manning coefficient based on hydraulic data, Eq. 2, varies 344 between 0.03 and 0.17 in the Desselse Nete. The empirically based Manning coefficients 345 16 overestimate this value every month up to a factor two. The largest differences are found in the 346 months May, June and August. 347 Discussion 348 There is a strong need for new methods to acquire 2D data on the spatial and temporal distribution 349 of vegetation in small rivers. The digital cover photography technique applied in this paper is a 350 useful tool to obtain this detailed 2D information. This method has six main advantages: (i) it can 351 be applied in rivers with any kind of vegetation cover; (ii) it has a very-high spatial resolution, 352 around 1.9 pixels cm-2 (~1.3 cm edge length), and a very flexible temporal resolution with the 353 frequency only dependent on availability of suitable weather conditions; (iii) it is relatively fast, 354 two to three days to collect and process the data of a reach of 180 m; (iv) it is non-destructive in 355 contrast to other methods where sampling is involved; (v) the equipment is relatively cheap with a 356 single time cost of approximately € 2000 for the camera, lens, control software and memory card; 357 (vi) the software used to process the data is widely available and similar open source alternatives 358 exist. Tab. 5 shows the performance of the current method in comparison to five other commonly 359 used remote sensing approaches with optical imagery. The spectral range and spectral resolution 360 depends on the sensor for all platforms mentioned in Tab. 5. Manned aircraft imaging can have a 361 wide range of spectral resolution from very narrow band hyperspectral imagery to one very broad 362 band for a panchromatic image. Similarly for satellite imagery, sensors with a high spectral 363 resolution are available. However, these images are of low spectral quality and low spatial 364 resolution. Hyperspectral sensors with a high spectral resolution are available for unmanned aerial 365 vehicles but can only be assembled on larger vehicles and do not achieve the high spatial resolution 366 that can be obtained with RGB cameras. The current method is particularly suitable for studies in 367 river reaches which are difficult to access and require high spatial resolution . In addition, limited 368 17 technical training is required to pre- and post-process the images. The method can be used in its 369 current stage in relative small study areas for monitoring, modelling and management purposes. 370 Applying this method in larger study areas would require further automatization of image 371 collection, e.g. by attaching the camera to an Unmanned Aerial Vehicle (UAV) (Husson et al. 2014; 372 Tamminga et al. 2015), and image classification, e.g. by applying the OBIA method (Visser et al. 373 2016). 374 The image data collection requires suitable light and site conditions. The water needs to be clear 375 (i.e. ideally < 1 m deep and low turbidity) (Visser et al. 2013), and the water velocity should be 376 low to limit stem motion (i.e. ideally < 1 ms-1) (Franklin et al. 2008). These site conditions are 377 similar requirements for the occurrence of macrophytes in the first place (Riis and Biggs 2003). 378 However the water can be temporary less clear after storm events. It this case it is recommended 379 to wait a few days until the concentration of suspended sediment is reduced. Light intensity should 380 be sufficient to penetrate the surface and illuminate the submerged macrophytes. The angle 381 between the sun and the camera should be around 45° to minimize sun glint and maximize the light 382 availability in the water. The time of image collection depends on the latitude of the study area, for 383 example in Belgium (latitude 52 °) this is around noon, between 11 a.m. and 1 p.m., summertime. 384 The image collection can only take place under these specified good weather conditions. This limits 385 the data collection frequency, but for monitoring vegetation very high frequency data is rarely 386 needed. Techniques currently under development may in the near future allow the removal of 387 remaining surface reflection (Hardesty 2015). Other requirements are related to the study area 388 itself. The rivers and streams should be relative small, i.e. <10 m wide, which is the equivalent of 389 the spatial extent covered by one image, and at least one river bank should be accessible and stable 390 enough to position the pole. Yet these limitations to the study area can be overcome by attaching 391 the camera to an Unmanned Aerial Vehicle (UAV) (Husson et al. 2014; Tamminga et al. 2015), or 392 18 to a helium balloon, or by attaching the pole to the bow of a boat (Lirman and Deangelo 2007). 393 This makes it possible to collect similar resolution data from close to the water surface of larger 394 rivers. However, helium balloons need to be sufficiently big to carry a DSLR camera, which makes 395 them rather impractical and in the long-run quite expensive platforms (due to the cost of helium). 396 UAVs are a good alternative since battery life is improving year on year. Currently the only 397 disadvantages of a rotary-winged UAV platform are (i) the need for training to actually fly the 398 vehicle, which may involve some costly training; (ii) the purchase and insurance of suitable quality 399 UAV and camera; (iii) the transport of larger UAVs. UAVs are therefore the platform of choice 400 for further development of the method proposed in this paper. 401 402 The image processing as it was done is this study works well, yet improvements are possible to 403 delineate and identify the vegetation patches. This study used a manual interpretation based on 404 expert judgement, which is a sound method to separate between different species (Husson et 405 al.2014), because the manual delineation and identification uses many image elements like size, 406 shape, shadow, colour, texture, pattern, location and surroundings (Colwell 1960; Tempfli et al. 407 2009). However, the observer bias can still be present since this method makes use of manual 408 decision rules concerning the exact edge of the vegetation patches. In the study reach dominated 409 by a single species the accuracy is very high (>97 %). These high values are due to the relative 410 simple composition of the vegetation patches, where the whole reach is covered by a single species 411 (Desselse Nete). On the contrary, the accuracy is less (> 59%) in the river with a heterogeneous 412 composition of multiple species (Zwarte Nete), certainly in months when the vegetation patches 413 are developing (June and July). If we compare the relative cover of each vegetation class between 414 the image method and ground survey, differences are less than 12 %. The images method proved 415 to be suitable to estimate the relative cover of each vegetation class in rivers with a continuous and 416 19 discontinuous vegetation cover. However, it is difficult to map the exact location of all vegetation 417 patches in rivers with heterogeneous vegetation cover. This is due to the movement of the 418 vegetation patches by the flowing water and the relatively simple image processing. 419 420 Another limitation is the detection of rare species which are normally not abundantl, e.g. C. 421 obtusangula was detected by the ground survey in June and July in the Zwarte Nete but not by the 422 image method. The last limitation is the separation of multi-layered plant communities, e.g. P. 423 natans was classified as S. emersum in August (Desselse Nete), while only a few leaves of S. 424 emersum where present on top of P. natans. Similar limitations are found by Anker et al. (2014). 425 From the images, plant growth form (submerged, submerged with floating leaves, emergent) can 426 be easily recognized, as well as the species identification up to genus level. A classification up to 427 species level is possible, but requires knowledge of the species present in the reach. This 428 information can simply be obtained during the collection of the images at the field site. Automatic 429 classification methods based on variation in spectral signatures of different vegetation types could 430 not be used to automatically delineate and identify vegetation patches under these specific 431 circumstances. The varying incidences of light, the prevailing sub-optimal light conditions during 432 the sampling campaign and submergence depth of the vegetation all caused complications for 433 automated species detection (Visser et al. 2013). We acknowledge this drawback on the manual 434 image processing, which increases the cost of data processing and may make this method no longer 435 as cost-effective. Attention should be given to reduce phenological (space and time) differences in 436 the classification to make this technique suitable for long term monitoring. Two solutions have 437 been proposed: (i) convert the Red-Green-Blue colors to the green chromatic coordinate 438 (G/[R + G + B]), (ii) use the 90th percentile of all daytime values within a three-day window around 439 the centre day (Dronova 2017; Sonnentag et al. 2012). However, it may be not straightforward to 440 20 apply similar algorithms to submerged aquatic vegetation where relative variation in Red-Green-441 Blue values at any point can differ due to water depth differences. Alternative image analysis 442 approaches such as object based image analysis (OBIA) are less reliant on spectral information and 443 may mitigate for such conditions, however applications of such approaches in submerged 444 environments are still in a developmental stage (Visser et al. 2016). OBIA is currently applied in 445 other ecosystems. For example Laba et al. (2010) used a maximum-likelihood classification in tidal 446 marshes, which resulted in a classification accuracy between 45 and 77 %. In offshore submerged 447 environments OBIA based approaches have so far achieved good results for mapping coarse 448 vegetation and substrate classes. For example, the extent of seagrass habitat was mapped by 449 Baumstark et al. (2016), showing in a slightly higher accuracy using OBIA (78%) compared to 450 photo-interpretation (71%). 451 The image analysis method proved suitable for measuring the spatio-temporal vegetation cover, 452 which is a primary parameter for monitoring vegetated ecosystems. For instance within the Water 453 Framework Directive, it is essential for long-term monitoring of vegetation abundance (Hering et 454 al. 2010). Changes in abundance and location of the vegetation were derived directly from the 455 image data. For example the regrowth capacity of P. natans was high after the mowing event in 456 June, and pre-mowed cover values were reached within 8 weeks, which is similar to other 457 macrophyte species (Bal et al. 2006). Other, more conventional methods to estimate vegetation 458 cover data range from fast methods with a high observer bias due to expert judgement (Tansley 459 scaling method based on 5 classes) to more detailed scaling methods, which have a higher accuracy, 460 but are more time consuming and require substantial expert knowledge (Braun-Blanquet scaling 461 method based on 9 classes (Blanquet 1928)) and Londo scaling method based on at least 21 classes 462 (Londo 1976)). These methods have two main disadvantages. Firstly, abundance class errors are 463 difficult to correct even with substantial expert knowledge (Wiederkehr et al. 2015). Secondly, the 464 21 classification of the cover makes use of discontinuous class scales, which are less accurate and can 465 hamper data analyses. Hence the image analysis method fulfils the requirement of a more objective 466 quantification of the cover with a continuous cover scale with high spatial and flexible temporal 467 resolution. 468 469 The cover maps were also used in this study to investigate the presence of spatial self-organisation 470 of macrophytes in lowland rivers. A significant power-law relationship of the frequency 471 distribution of the patch sizes is found, which is an indication of spatial self-organisation (Newman 472 2005; Scanlon et al. 2007). This is in agreement with a study of Schoelynck et al. (2012), who 473 investigated the spatial self-organisation of macrophytes in the same reach in the Zwarte Nete in 474 2008. In the study of Schoelynck et al. (2012) the exact location of all vegetation patches was 475 determined using an electronic theodolite. It took roughly three weeks to map the whole reach, 476 which is much slower in comparison with the new method, were we needed 1 hour to collect the 477 images and two to three days to process the data. So obtaining spatial information of vegetation is 478 much faster compared to conventional methods. 479 480 From the cover data, biomass can be derived using simple non-destructive cover:biomass 481 conversion factors. These conversion factors can be determined for the specific field site or can be 482 obtained from literature (e.g. Madsen and Adams (1989); Flynn et al. (2002)). The biomass (gDW 483 m-2) estimated by the image analysis method was compared to the biomass obtained from the 484 scheduled mowing method. The biomass obtained by the two methods does not significantly differ 485 for either of the two reaches. The relatively small differences may be attributed to inaccuracies in 486 both methods. During the scheduled mowing, the biomass could have been slightly overestimated 487 when non-plant materials like sediment, stones and dead wood were removed too, which may have 488 22 added up to the total fresh weight, or underestimated the latter when not all the vegetation was 489 removed. However, we only assessed the biomass in a month with high biomass. Higher relative 490 difference in biomass might be expected when less biomass is present, but this would result in low 491 absolute differences. The image analysis method may also have certain flaws and uncertainties 492 involving the estimation of the species-specific biomass obtained by the plots. The within species 493 variation of the biomass may not be fully captured by three replicas (e.g. by depth variance of the 494 river and of the vegetation). The image analysis method doesn’t account for variability in the 495 density. Classic methods of biomass estimation are based on destructive measures of the biomass 496 (mowing, harvesting), which disturb the follow-up of natural vegetation development during the 497 growth season (Wood et al. 2012). 498 499 The difference between the Manning coefficient based on empirical relationships and the one based 500 on hydraulic data differs less than 23 % in the Zwarte Nete and less than 37% in the Desselse Nete. 501 The empirical relationships don’t account for the species composition and horizontal and vertical 502 distribution of the vegetation, which are different in both rivers and are major determining factors 503 of the hydraulic resistance of the reach. The Zwarte Nete is dominated by submerged vegetation 504 and this vegetation type has similar effects on the hydraulic resistance as the vegetation used to 505 construct Eq. 2 and Eq. 3. The Desselse Nete is dominated by the floating species P. natans, which 506 is a more open species that concentrates the majority of the biomass near the water surface, which 507 leads to a limited interaction with the water flow: rivers with macrophytes can have a 2 to 7-fold 508 increase of the resistance for floating (Green 2005a) and submerged (Bal and Meire 2009) species, 509 respectively, compared to rivers without vegetation. The same vegetation biomass or cover will 510 therefore result in a lower hydraulic resistance. Detailed 2D hydrodynamic models can be used to 511 quantify more accurately the hydraulic resistance created by the vegetation based on plant density, 512 23 species characteristics and spatial distribution of the vegetation (Verschoren et al. 2015). Accurate 513 2D spatio-temporal vegetation cover data, as obtained by the digital cover photography technique, 514 is indispensable to calibrate and validate these models. The spatial distribution of the vegetation is 515 a direct input to these models. Therefore these models account for the exact location of all 516 vegetation patches and the different plant characteristics of all species. This is a major leap forward 517 for engineers and water managers in the fine tuning of the hydrodynamic models of vegetated 518 rivers. 519 24 Conclusions 520 We successfully applied a digital cover photography technique based on orthogonal aerial images 521 with a very-high spatial (subcentimetre) and flexible temporal (monthly) resolution. The produced 522 vegetation maps were used to assess four key properties of vegetated lowland rivers which are 523 important for monitoring, modelling and management, being spatio-temporal variation in 524 vegetation cover, patch size distribution, biomass and hydraulic resistance. 525 The main limitations are related to the study area itself, which should be limited in size, and the 526 prevailing light conditions should be sufficient to look into the water. Improvements in the images 527 processing are situated in the automatic delineation and classification of the vegetation patches. 528 529 25 Acknowledgements 530 The funding for this research was partly provided by the Research Fund Flanders (FWO-, project 531 no. G.0290.10) via the multidisciplinary research project ‘Linking optical imaging techniques and 532 2D-modelling for studying spatial heterogeneity in vegetated streams and rivers’ (University of 533 Antwerp and University of Ghent) and party by Province of Antwerp, departement Leefmilieu, 534 dienst Integraal Waterbeleid (Report number ECOBE – 014 – R179). V.V. thanks the Institute for 535 the Promotion of Innovation through Science and Technology in Flanders (IWT-Vlaanderen) for 536 personal research funding. J.S. is a postdoctoral fellow of FWO (project no. 12H8616N). 537 538 References 539 Anker, Y., Y. Hershkovitz, E. Ben Dor & A. Gasith, 2014. Application of Aerial Digital Photography 540 for Macrophyte Cover and Composition Survey in Small Rural Streams. River Res Appl 541 30(7):925-937. 542 Apollo Mapping, 2016. RapidEye. In. https://apollomapping.com/imagery/medium-resolution-543 satellite-imagery. 544 Bal, K. D. & P. Meire, 2009. The Influence of Macrophyte Cutting on the Hydraulic Resistance of 545 Lowland Rivers. J Aquat Plant Manage 47:65-68. 546 Bal, K. D., E. Struyf, H. Vereecken, P. Viaene, L. De Doncker, E. de Deckere, F. Mostaert & P. 547 Meire, 2011. How do macrophyte distribution patterns affect hydraulic resistances? Ecol 548 Eng 37(3):529-533. 549 Bal, K. D., S. Van Belleghem, E. De Deckere & P. Meire, 2006. The re-growth capacity of sago 550 pondweed following mechanical cutting. J Aquat Plant Manage 44:139-141. 551 Barrell, J. & J. Grant, 2015. High-resolution, low-altitude aerial photography in physical geography: 552 A case study characterizing eelgrass (Zostera marina L.) and blue mussel (Mytilus edulis 553 L.) landscape mosaic structure. Prog Phys Geog 39(4):440-459. 554 Battle, J. M. & T. B. Mihuc, 2000. Decomposition dynamics of aquatic macrophytes in the lower 555 Atchafalaya, a large floodplain river. Hydrobiologia 418:123-136. Blanquet, B., 1928. 556 Pflanzensoziologie. Grundzüge der Vegegtationskunde, Berlin 1.ed. Berlin. 557 Baumstark R., R. Duffey & R. Pu, 2016. Mapping seagrass and colonized hard bottom in Springs 558 Coast, Florida using WorldView-2 satellite imagery. Estuarine, Coastal and Shelf Science 559 181:83-92.Borin, M. & M. Salvato, 2012. Effects of five macrophytes on nitrogen 560 remediation and mass balance in wetland mesocosms. Ecol Eng 46:34-42. 561 Bryson, M., M. Johnson-Roberson, R. J. Murphy & D. Bongiorno, 2013. Kite Aerial Photography 562 for Low-Cost, Ultra-high Spatial Resolution Multi-Spectral Mapping of Intertidal 563 Landscapes. Plos One 8(9). 564 Carter, G. A., A. K. Knapp, J. E. Anderson, G. A. Hoch & M. D. Smith, 2005. Indicators of plant 565 species richness in AVIRIS spectra of a mesic grassland. Remote Sens Environ 98(2-566 3):304-316. 567 http://www.sciencedirect.com/science/journal/02727714 26 Champion, P. D. & C. C. Tanner, 2000. Seasonality of macrophytes and interaction with flow in a 568 New Zealand lowland stream. Hydrobiologia 441(1-3):1-12. 569 Chow, V. T., 1959. Open-channel hydraulics. McGraw-Hill, New York, USA. 570 Colwell, R., 1960. Manual of photographic interpretation Washington, DC. 571 Cordova, J. M., E. J. Rosi-Marshall, J. L. Tank & G. A. Lamberti, 2008. Coarse particulate organic 572 matter transport in low-gradient streams of the Upper Peninsula of Michigan. J N Am 573 Benthol Soc 27(3):760-771. 574 De Doncker, L., P. Troch, R. Verhoeven, K. Bal, N. Desmet & P. Meire, 2009a. Relation between 575 Resistance Characteristics Due to Aquatic Weed Growth and the Hydraulic Capacity of the 576 River Aa. River Res Appl 25(10):1287-1303. 577 De Doncker, L., P. Troch, R. Verhoeven, K. Bal, P. Meire & J. Quintelier, 2009b. Determination of 578 the Manning roughness coefficient influenced by vegetation in the river Aa and Biebrza 579 river. Environ Fluid Mech 9(5):549-567. 580 Dhote, S. & S. Dixit, 2009. Water quality improvement through macrophytes-a review. Environ 581 Monit Assess 152(1-4):149-153. 582 Dinka, M., E. Agoston-Szabo & I. Toth, 2004. Changes in nutrient and fibre content of 583 decomposing Phragmites australis litter. Int Rev Hydrobiol 89(5-6):519-535. 584 Dronova, I. 2017. Environmental heterogeneity as a bridge between ecosystem service and visual 585 quality objectives in management, planning and design. Landscape and Urban Planning 586 163:90-106. 587 EU, 2000. 2000/60/EC. A framework for the Community action in the field of water policy, or 588 short: The EU Water Framework Directive Directive 2000/60/EC, The European Parliament 589 and the European Council of Ministers. Brussels, Belgium. 590 Extence, C. A., D. M. Balbi & R. P. Chadd, 1999. River flow indexing using British benthic 591 macroinvertebrates: A framework for setting hydroecological objectives. Regul River 592 15(6):543-574. 593 Flynn, K. F. & S. C. Chapra, 2014. Remote Sensing of Submerged Aquatic Vegetation in a Shallow 594 Non-Turbid River Using an Unmanned Aerial Vehicle. Remote Sens-Basel 6(12):12815-595 12836. 596 Flynn, N. J., D. L. Snook, A. J. Wade & H. P. Jarvie, 2002. Macrophyte and periphyton dynamics 597 in a UK Cretaceous chalk stream: the River Kennet, a tributary of the Thames. Sci Total 598 Environ 282:143-157. 599 Foody, G. M., 2002. Status of land cover classification accuracy assessment. Remote Sens Environ 600 80(1):185-201. 601 Franklin, P., M., Dunbar & P. Whitehead, 2008. Flow controls on lowland river macrophytes: A 602 review. Sci Total Environ 400(1-3):369-378. 603 Gibbins, C., D. Vericat & R. J. Batalla, 2007. When is stream invertebrate drift catastrophic? The 604 role of hydraulics and sediment transport in initiating drift during flood events. Freshwater 605 Biol 52(12):2369-2384. 606 Green, J. C., 2005a. Comparison of blockage factors in modelling the resistance of channels 607 containing submerged macrophytes. River Res Appl 21(6):671-686. 608 Green, J. C., 2005b. Modelling flow resistance in vegetated streams: review and development of 609 new theory. Hydrol Process 19(6):1245-1259. 610 Guichard, F., E. Bourget & J. P. Agnard, 2000. High-resolution remote sensing of intertidal 611 ecosystems: A low-cost technique to link scale-dependent patterns and processes. Limnol 612 Oceanogr 45(2):328-338. 613 27 Gurnell, A. M., M. P. van Oosterhout, B. de Vlieger & J. M. Goodson, 2006. Reach-scale interactions 614 between aquatic plants and physical habitat: River Frome, Dorset. River Res Appl 615 22(6):667-680. 616 Hardesty, L., 2015. Remowing reflections from photos taken through windows: new algorithm 617 exploits multiple reflections in individual images to distinguish reflection from transmission. 618 In: Massachusetts Institute of Technology. 619 Hering, D., A. Borja, J. Carstensen, L. Carvalho, M. Elliott, C. K. Feld, A. S. Heiskanen, R. K. 620 Johnson, J. Moe, D. Pont, A. L. Solheim & W. Van De Bund, 2010. The European Water 621 Framework Directive at the age of 10: A critical review of the achievements with 622 recommendations for the future. Sci Total Environ 408(19):4007-4019. 623 Hestir, E. L., S. Khanna, M. E. Andrew, M. J. Santos, J. H. Viers, J. A. Greenberg, S. S. Rajapakse 624 & S. L. Ustin, 2008. Identification of invasive vegetation using hyperspectral remote 625 sensing in the California Delta ecosystem. Remote Sens Environ 112(11):4034-4047. 626 Horvath, T. G., 2004. Retention of particulate matter by macrophytes in a first-order stream. 627 Aquat Bot 78(1):27-36. 628 Husson, E., O. Hagner & F. Ecke, 2014. Unmanned aircraft systems help to map aquatic 629 vegetation. Appl Veg Sci 17(3):567-577. 630 Krause, S., D. M. Hannah, J. H. Fleckenstein, C. M. Heppell, D. Kaeser, R. Pickup, G. Pinay, A. L. 631 Robertson & P. J. Wood, 2011. Inter-disciplinary perspectives on processes in the 632 hyporheic zone. Ecohydrology 4(4):481-499. 633 Lamberti, G. A., S. V. Gregory, L. R. Ashkenas, R. C. Wildman & A. D. Steinman, 1989. Influence 634 of Channel Geomorphology on Retention of Dissolved and Particulate Matter in a Cascade 635 Mountain Stream. Us for Serv T R Psw 110:33-39. 636 Laba, M., B. Blair, R. Downs, B. Monger, W. Philpot, S. Smith, P. Sullivan & P. C. Baveye, 2010. 637 Use of textural measurements to map invasive wetland plants in the Hudson River National 638 Estuarine Research Reserve with IKONOS satellite imagery. Remote Sensing of 639 Environment 114:876-886. 640 Legleiter, C. J., 2003. Spectrally driven classification of high spatial resolution, hyperspectral 641 imagery: A tool for mapping in-stream habitat. Environ Manage 32(3):399-411. 642 Lejeune, O., M. Tlidi & R. Lefever, 2004. Vegetation spots and stripes: Dissipative structures in 643 arid landscapes. Int J Quantum Chem 98(2):261-271. 644 Lirman, D. & G. Deangelo, 2007. Geospatial video monitoring of benthic habitats using the 645 Shallow-Water Positioning System (SWaPS). Oceans-Ieee:1176-1180. 646 Londo, G., 1976. The decimal scale for releves of permanent quadrats, vol 33. 647 Madsen, J. D. & M. S. Adams, 1989. The Distribution of Submerged Aquatic Macrophyte Biomass 648 in a Eutrophic Stream, Badfish Creek - the Effect of Environment. Hydrobiologia 649 171(2):111-119. 650 Madsen, J. D., P. A. Chambers, W. F. James, E. W. Koch & D. F. Westlake, 2001. The interaction 651 between water movement, sediment dynamics and submersed macrophytes. 652 Hydrobiologia 444(1-3):71-84. 653 Masek, J. G., D. J. Hayes, M. J. Hughes, S. P. Healey & D. P. Turner, 2015. The role of remote 654 sensing in process-scaling studies of managed forest ecosystems. Forest Ecol Manag 655 355:109-123. 656 Menendez, M., O. Hernandez & F. A. Comin, 2003. Seasonal comparisons of leaf processing rates 657 in two Mediterranean rivers with different nutrient availability. Hydrobiologia 495(1-3):159-658 169. 659 28 Mitchell, J. J., R. Shrestha, L. P. Spaete & N. F. Glenn, 2015. Combining airborne hyperspectral 660 and LiDAR data across local sites for upscaling shrubland structural information: Lessons 661 for HyspIRI. Remote Sens Environ 167:98-110. 662 NASA, 2016. MODIS, Moderate Resolution Imaging Spectroradiometer. In. 663 http://modis.gsfc.nasa.gov. 664 Newman, M. E. J., 2005. Power laws, Pareto distributions and Zipf's law. Contemporary Physics 665 46(5):323-351. 666 Nezlin, N. P., K. Kamer & E. D. Stein, 2007. Application of color infrared aerial photography to 667 assess macroalgal distribution in an eutrophic estuary, upper Newport Bay, California. 668 Estuar Coast 30(5):855-868. 669 Pekin, B. & C. Macfarlane, 2009. Measurement of Crown Cover and Leaf Area Index Using Digital 670 Cover Photography and Its Application to Remote Sensing. Remote Sens-Basel 1(4):1298-671 1320. 672 R Core Team, 2014. R: A language and environment for statistical computing. 673 Rango, A., A. Laliberte, K. Havstad, C. Winters, C. Steele & D. Browning, 2010. Rangeland 674 Resource Assessment, Monitoring, and Management Using Unmanned Aerial Vehicle-675 Based Remote Sensing. Int Geosci Remote Se:608-611. 676 Rietkerk, M., S. C. Dekker, P. C. de Ruiter & J. van de Koppel, 2004. Self-organized patchiness 677 and catastrophic shifts in ecosystems. Science 305(5692):1926-1929. 678 Riis, T. & B. J. F. Biggs, 2003. Hydrologic and hydraulic control of macrophyte establishment and 679 performance in streams. Limnol Oceanogr 48(4):1488-1497. 680 Sand-Jensen, K., K. Andersen & T. Andersen, 1999. Dynamic properties of recruitment, expansion 681 and mortality of macrophyte patches in streams. Int Rev Hydrobiol 84(5):497-508. 682 Satellite Imaging Corporation, 2016. Ikonos Satellite Sensor. In. 683 http://www.satimagingcorp.com/satellite-sensors/ikonos/. 684 Scanlon, T. M., K. K. Caylor, S. A. Levin & I. Rodriguez-Iturbe, 2007. Positive feedbacks promote 685 power-law clustering of Kalahari vegetation. Nature 449(7159):209-U4. 686 Schoelynck, J., T. De Groote, K. Bal, W. Vandenbruwaene, P. Meire & S. Temmerman, 2012. Self-687 organised patchiness and scale-dependent bio-geomorphic feedbacks in aquatic river 688 vegetation. Ecography 35(8):760-768. 689 Seitzinger, S., J. A. Harrison, J. K. Bohlke, A. F. Bouwman, R. Lowrance, B. Peterson, C. Tobias & 690 G. Van Drecht, 2006. Denitrification across landscapes and waterscapes: A synthesis. Ecol 691 Appl 16(6):2064-2090. 692 Shuchman, R. A., M. J. Sayers & C. N. Brooks, 2013. Mapping and monitoring the extent of 693 submerged aquatic vegetation in the Laurentian Great Lakes with multi-scale satellite 694 remote sensing. J Great Lakes Res 39:78-89. 695 Silva, T. S. F., M. P. F. Costa, J. M. Melack & E. M. L. M. Novo, 2008. Remote sensing of aquatic 696 vegetation: theory and applications. Environ Monit Assess 140(1-3):131-145. 697 Sonnentag, O., Hufkens, K., Teshera-Sterne, C., Young, A.M., Friedl, M.A., Braswell, B.H., Miliman, 698 T., O'Keefe, J., Richardson, A.D. (2012), Digital repeat photography for phenological 699 research in forest ecosystems. Agricultural and Forest Meteorology, 152, 159-177. 700 Tamminga, A., C. Hugenholtz, B. Eaton & M. Lapointe, 2015. Hyperspatial Remote Sensing of 701 Channel Reach Morphology and Hydraulic Fish Habitat Using an Unmanned Aerial Vehicle 702 (Uav): A First Assessment in the Context of River Research and Management. River Res 703 Appl 31(3):379-391. 704 Tempfli, K., N. Kerle, G. Huurneman & L. Janssen, 2009. Principles of remote sensing, Enschede, 705 NL. 706 29 U.S. Department of the Interior & U.S. Geological Survey, 2016. Landsat Missions. In. 707 http://landsat.usgs.gov. 708 van de Koppel, J., J. C. Gascoigne, G. Theraulaz, M. Rietkerk, W. M. Mooij & P. M. J. Herman, 709 2008. Experimental Evidence for Spatial Self-Organization and Its Emergent Effects in 710 Mussel Bed Ecosystems. Science 322(5902):739-742. 711 van der Wal, D., J. van Dalen, A. Wielemaker-van den Dool, J. T. Dijkstra & T. Ysebaert, 2014. 712 Biophisical contral of the intertidal benthic macroalgae revealed by high-frequency 713 multispectral camera images. J Sea Res 90:111-120. 714 Vereecken, H., J. Baetens, P. Viaene, F. Mostaert & P. Meire, 2006. Ecological management of 715 aquatic plants: effects in lowland streams. Hydrobiologia 570:205-210. 716 Verschoren, V., D. Meire, J. Schoelynck, K. Buis, K. D. Bal, P. Troch, P. Meire & S. Temmerman, 717 2016. Resistance and reconfiguration of natural flexible submerged vegetation in 718 hydrodynamic river modelling Environ Fluid Mech 16(1), 245-265. 719 Villa, P., M. Bresciani, R. Bolpagni, M. Pinardi & C. Giardino, 2015. A rule-based approach for 720 mapping macrophyte communities using multi-temporal aquatic vegetation indices. 721 Remote Sens Environ 171:218-233. 722 Visser, F., C. Wallis & A. M. Sinnott, 2013. Optical remote sensing of submerged aquatic 723 vegetation: Opportunities for shallow clearwater streams. Limnologica 43(5):388-398. 724 Visser, F., K. Buis, V. Verschoren & J. Schoelynck, 2016. Mapping of submerged aquatic in rivers 725 from very high-resolution image data, using object-based image analysis combined with 726 expert knowledge. Hydrobiologia:19 doi:10.1007/s10750-016-2928-y. 727 Wiederkehr, J., C. Grac, M. Fabregue, B. Fontan, F. Labat, F. Le Ber & M. Tremolieres, 2015. 728 Experimental study of uncertainties on the macrophyte index (IBMR) based on species 729 identification and cover. Ecol Indic 50:242-250. 730 Wood, K. A., R. A. Stillman, R. T. Clarke, F. Daunt & M. T. O'Hare, 2012. Measuring submerged 731 macrophyte standing crop in shallow rivers: A test of methodology. Aquat Bot 102:28-33. 732 30 Tables 733 Table 1: Overview of the equations used to calculate the Manning coefficient, n (s m-1/3). Eq. 2 is used to calculate the 734 Manning coefficient, with A (m2) cross-sectional area, Q (m3 s-1) discharge, R (m) hydraulic radius, S (m m-1) water 735 level slope, for which all parameters are measured in both reaches of the study area. Eq. 3 and Eq. 4 are empirical 736 relationships between the Manning coefficient and the vegetation cover (%) and Manning coefficient and the biomass 737 (g DW m-2), respectively All parameters are derived from the digital maps. 738 Reference Equation Number Chow et al. (1956) 𝑛 = 𝐴 𝑄 ∗ 𝑅2 3⁄ ∗ 𝑆1 2⁄ Eq. 2 Green (2005) 𝑛 = 0.0438 exp (0.0200 ∗ 𝑐𝑜𝑣𝑒𝑟) Eq. 3 De Doncker et al. (2009) 𝑛 = 0.4628 − 0.3998 exp (−0.0047 ∗ 𝑏𝑖𝑜𝑚𝑎𝑠𝑠) Eq. 4 739 740 31 Table 2: The accuracy (%) of the species identification of the image method compared to the ground survey method 741 per month per river. The accuracy is based on species level; for each grid cell (n=396) the species is compared between 742 the image method and the ground survey method. 743 Month April May June July August September Zwarte Nete 100 100 66.4 59.6 84.8 93.7 Desselse Nete 100 100 100 100 97.0 100 744 745 32 Table 3:Percentage vegetation cover (%) estimated by the image method and the ground survey method (GS) for June, 746 July, August and September in the Zwarte Nete. 747 Month June July August September Method GS Image GS Image GS Image GS Image C. obtusangula 2.3 0.0 2.0 0.0 2.5 2.5 0.0 0.0 M. spicatum - - - - - - - - P. pectinatus 1.3 0.0 25.5 24.0 5.1 0.0 - - S. emersum 32.3 29.6 62.1 59.3 86.4 97.5 4.6 6.1 Riparian vegetation - - - - - - - - Bare sediment 64.1 70.4 10.4 16.6 1.0 0.0 95.5 94.0 748 33 Table 4:Total biomass (gDW m-2) per month in both rivers. The biomass is estimated by the image analysis method 749 and by mowing method when all vegetation was removed and weighed. 750 Month April May June July August September Zwarte Nete Image 0 3.3 ± 0.1 11.4 ± 3.0 56.8 ± 5.9 187.5 ± 34.2 10.8 ± 1.3 Mowing - - - - 193.3 - Desselse Nete Image 0.9 65.8 ± 8.8 101.6 ± 26.4 36.7 ± 7.9 150.4 ± 24.4 13.8 ± 3.5 Mowing - - - - 123.6 - 751 34 Table 5: Comparison of the current method with fiveother remote sensing approaches using optical imagery and ground level visual survey. The features where the 752 current method performs good are highlighted in bold. 753 Spatial resolution (pixel edge length) Temporal resolution Spectral region Operation cost Collection cost Spatial extent Weather dependency Knowledge requirements (obtaining, processing) This study < 1 cm Flexible RGB Low (man hours, consumables) Low m² Low (sun) Low Kite, blimp and balloon photography (Barrell and Grant 2015; Bryson et al. 2013; Guichard et al. 2000) < 5 cm Flexible RBG NIR Low (man hours, consumables) Medium m² Medium (sun, wind speed) Medium Unmanned aerial vehicles (Rango et al. 2010) 1-10 cm (dependent on sensor and flight height) Flexible RBG NIR High (training, man hours, post-processing) Medium m² - hm² Medium (sun, wind speed) High Manned aircraft imaging 0.3 - 5m (dependent on flying height) Flexible RGB NIR MIR High (plane charter, post-processing) High m² - km² High (sun, sky conditions) High Freely available satellite images (NASA 2016; U.S. Department of the Interior and U.S. Geological Survey 2016) > 5 m Fixed Several/year (dependent on location and resolution) RGB NIR MIR 0 0 > 1 km² High (sun, sky conditions) Medium Commercial satellite images (Apollo Mapping 2016; Satellite Imaging Corporation 2016) 0.5-5 m Fixed 14-100 days (dependent on location and resolution) RGB NIR MIR 0 High > 1 km² High (sun, sky conditions) Medium Ground level visual survey Variable Flexible - High - m² None Low 754 35 Figures 755 756 Figure 1: The location of the study area is indicated with a black dot in the North East of Belgium. Insert: the location 757 of Belgium in Europa is shown in dark grey. 758 759 36 760 (a) (b) Figure 2: Illustrations of the image collection in the field. (a) The DSLR camera is attached with a ball head to a 761 handheld telescopic pole to take orthogonal images. (b) One person holds the pole with camera tilted in order to 762 position the camera at a height of 5 m above the water surface. A second person checks with a live view on a laptop 763 that both river banks are visible on each image and takes the images with tethered capture. 764 765 37 (a) (b) (c) (d) (e) (f) → → ↓ → ↓ → 38 Figure 3: Examples are given of the image collection, processing and analysis in the (a, c, e) Zwarte Nete and the (b, 766 d, e) Desselse Nete on the 13th of August 2013. Illustrations are shown of (a, b) individual images taken with a DSLR 767 camera attached to a pole, (c, d) a plan view of a part of the image mosaic, (e, f) vegetation map with colors indicating 768 the species and the location of the ground survey (black rectangular). The water flow direction is indicated with an 769 arrow. 770 771 39 772 773 774 Figure 4: Vegetation cover per species per month for the reach in the (a) Zwarte Nete and (b) Desselse Nete. The 775 colors of the bars refer to the species, the same colors for the species as in Fig. 3 are used (submerged species: red-776 yellow, floating species: green, emerged species: blue). The total vegetation cover per month is added in italics. 777 778 40 779 Figure 5: The inverse cumulative distribution of the patch sizes of C. obtusangula plotted on a double logarithmic 780 scale. A power-law relationship is added with β = 0.6 of Eq.1 (p<0.001; R² = 0.99). 781 782 41 783 784 Figure 6: Manning coefficient in function of time for the (a) Zwarte Nete and (b) Desselse Nete. For the validation 785 the Manning coefficient is calculated with Eq. 2() based on field measurents, Table 1. The Manning coefficient is 786 calculated with Eq. 3 ( ) and Eq. 4 (), these are empirical relationships with cross-sectional blockage and biomass, 787 respectively see Table 1. 788 789 https://www.google.be/url?sa=i&rct=j&q=&esrc=s&source=images&cd=&cad=rja&uact=8&ved=0CAcQjRxqFQoTCNPXzJW67cgCFYbTFAodZGMLDw&url=https://www.knutselkraam.nl/papier-karton/vierkant-karton-27-x-135-cm-hsg-220-gr/vierkant-donker-grijs-220-gr/&psig=AFQjCNF6KWoadENbhQYIQe7V-LOuPvyvvw&ust=1446406477967479 42 Appendix 790 Table 6: Overview of the measured hydraulic data per river per month. These values are used to calculate the Manning 791 coefficient with Eq. 2. 792 April May June July August September Zwarte Nete Discharge (m3 s-1) 0.28 0.5 0.23 0.25 0.2 0.46 Cross-sectional area (m2) 1.06 1.39 1.28 1.97 2.36 1.69 Hydraulic radius (m) 0.35 0.43 0.37 0.43 0.51 0.39 Water level slope (m m-1) 0.0007 0.0007 0.0009 0.0012 0.0013 0.0005 Desselse Nete Discharge (m3 s-1) 0.45 0.61 0.33 0.39 0.32 0.61 Cross-sectional area (m2) 1.43 1.63 1.76 1.90 2.65 2.32 Hydraulic radius (m) 0.38 0.33 0.44 0.46 0.57 0.52 Water level slope (m m-1) - 0.0005 0.0009 0.0006 0.0009 0.0008 793 Table 7: The biomass:cover conversion factor mean ± standard error (g m-2) is measured per month for C. 794 obtusangula, S. emersum and P. natans (n=3). Note that no replicates were taken in April, so no standard error is 795 given. 796 Apr. May Jun. Jul. Aug. Sept. C. obtusangula 28.5 NA 114.8 ± 37.7 123.7 ± 8.6 238.4 ± 50.3 354.9 ± 41.0 P. natans 146.0 116.6 ± 15.9 172.8 ± 45.3 209.8 ± 48.3 174.5 ± 19.5 190.6 ± 67.6 S. emersum 1.1 4.6 ± 1.7 49.9 ± 8.5 85.3 ± 14.0 202.2 ± 62.0 64.0 ± 10.7 797