Assessing intertidal sediment photopigment content from spectral reflectance 1 with an UAV-mounted 10-band multispectral sensor 2 3 4 Tristan J. Douglasa, Nicholas C. Coopsa, Mark C. Dreverb 5 6 a Faculty of Forestry, 2424 Main Mall, University of British Columbia, Vancouver, BC V6T 7 1Z4, Canada 8 b Environment & Climate Change Canada, Pacific Wildlife Research Centre, 5421 Robertson 9 Road, Delta, BC V4K 3N2, Canada 10 11 12 AUTHORS' CONTRIBUTIONS 13 T.J.D., N.C.C., and M.C.D. conceptualized the study and designed the methodology; T.J.D. 14 conducted fieldwork, laboratory analysis, and data processing; T.J.D. wrote the original 15 manuscript draft; T.J.D., N.C.C., and M.C.D. contributed critically to the drafts and gave final 16 approval for publication. 17 18 ACKNOWLEDGEMENT 19 We thank Professor Maria T. Maldonado, Dr. Jian Guo, and Maureen Soon for generously 20 offering their geochemical analysis expertise and providing access to their laboratory facilities 21 for photopigment analysis. 22 23 FUNDING INFORMATION 24 This work was supported by Funding was provided by the Environment and Climate Change 25 Canada [grant number 3000751679]. 26 27 CONFLICT OF INTEREST 28 The authors declare that they have no known competing financial interests or personal 29 relationships that could have appeared to influence the work reported in this paper. 30 31 Abstract 32 1. Remotely sensed multispectral information can be leveraged to quantify microphytobenthic 33 (MPB) biofilm biomass in intertidal mudflats and overcome some of the limitations of 34 traditional field campaigns with sediment collection. Satellite-based approaches are well 35 established, however moderate spatial resolution (≥ 10 m) is not suitable for investigating 36 fine-scale MPB spatial heterogeneity. Multispectral sensors mounted to unoccupied aerial 37 vehicles (UAVs) can fill this gap by providing data products with cm-scale pixels, but no 38 standard protocol currently exists for calibrating UAV-acquired spectral information to 39 sediment MPB content. 40 2. Here, we present a new protocol for calibrating data from a UAV-mounted multispectral 41 sensor (MicaSense RedEdge-MX dual) to sediment MPB biomass as measured by 42 photopigment content. To do so, we (1) built an adaptable methodology for acquiring and 43 analyzing UAV imagery and sediment photopigment field data, then (2) implemented a 44 protocol in the Fraser River Estuary, Canada, to build a statistically valid calibration 45 equation, testing the effectiveness of several spectral indices and photopigment 46 measurements. 47 3. Calibrated spectral index values from MicaSense RedEdge-MX data can provide a very 48 accurate measurement of MPB biomass, able to achieve 90 % correlation between the 49 normalized difference vegetation index (NDVI) and sediment chlorophyl-a (chl-a) 50 concentration. This high performance was achieved by (1) closely pairing georeferenced 51 sediment samples to corresponding multispectral imagery, (2) selecting an accessible section 52 of mudflats with a wide range of sediment photopigment concentrations, and (3) minimizing 53 the lag between sediment sample collection and MicaSense imagery acquisition. 54 4. This newly developed protocol can facilitate the use of calibrated UAV-acquired MicaSense 55 RedEdge-MX multispectral imagery of mudflats for investigating ecologically-important 56 fine-scale spatial heterogeneity and short-term temporal dynamics of MPB biomass, and can 57 bridge the gap between UAV- and satellite- acquired imagery of intertidal ecosystems. 58 59 Keywords: Microphytobenthos, MPB, biofilm, intertidal, mudflats, sediment, estuary, Fraser 60 River, fluorometry, chlorophyll-a, chl-a, phaeopigments, biogeochemistry, UAV, UAS, RPAS, 61 drone, photogrammetry, structure from motion, SfM, multispectral, micasense, reflectance, 62 remote sensing, calibration, modelling 63 1. Introduction 64 Microphytobenthic (MPB) biofilms contribute significantly to the biogeochemistry of intertidal 65 ecosystems, facilitating sediment cohesion, nutrient and oxygen cycling, carbon sequestration, 66 and provision of food resources for higher trophic levels (Hope et al., 2020). The biomass, 67 spatial distribution, and temporal dynamics of the MPB are highly variable, making data derived 68 from traditional field campaigns with sediment collection prone to high variability with limited 69 validity beyond the sampling location (Jacobs et al., 2021). In recent decades, satellite-acquired 70 multispectral information has been increasingly leveraged to map MPB biomass across broad 71 spatial extents with a minimum spatial resolution of 10 m (Daggers et al., 2018; Méléder et al., 72 2020; Oiry & Barillé, 2021). However, finer-scale MPB mapping is also necessary to both 73 resolve issues of spectral mixing within pixels and to better understand the influence of fine-74 scale MPB spatial heterogeneity on ecological processes. For these purposes, multispectral 75 sensors mounted to unoccupied aerial vehicles (UAVs) have proven to be a promising, cost-76 effective solution, offering the potential for frequent collection of imagery with cm-scale spatial 77 resolutions (Brunier et al., 2022; Douglas et al., 2023). Because these technologies are so novel, 78 however, calibration models relating UAV- acquired spectral information and sediment MPB 79 content have not been as well established as those derived from satellite imagery. 80 Both chemical analysis of sediment and spectroradiometric techniques rely the properties 81 of photopigments to determine MPB biomass (Pinckney et al., 2008). Chlorophyll-a (chl-a) is the 82 only pigment found in all plants, algae, and cyanobacteria and is a well-established proxy for 83 microphytobenthos biomass within intertidal sediments (MacIntyre et al., 1996). Chl-a extracted 84 from sediment is quantified by either fluorometry, spectrophotometry, or chromatography. For 85 remotely-sensed data, MPB biomass is estimated by the optical properties of photopigments, 86 which absorb downwelling, photosynthetically active radiation (PAR) at wavelengths ranging 87 from 300 to 800 nm and transfer it into chemical energy (Méléder et al., 2010). Chl-a has 88 absorption peaks around 440 and 675 nm in the blue and red regions of electromagnetic 89 radiation, respectively. Photon energy in the near-infrared spectral region (0.7 to 3.0 µm) is 90 emitted from healthy vegetation, as it is too small to synthesize organic molecules and can cause 91 cellular damage. Thus, the fraction of absorbed light and reflected light depends on the pigment 92 composition and the relative and quantitative abundance (biomass) of microphytobenthic 93 biofilm. 94 Spectral indices derived from acquired radiometric data are designed to assess the 95 contribution of photosynthetic pigments and vegetation biomass properties and can be used to 96 map MPB biomass. These indices can be calculated as either single-band ratios or normalized 97 differences from near infrared (NIR) and visible bands (Méléder et al., 2010). Empirical 98 relationships between a number of these spectral indices and chl-a have been tested and validated 99 for mapping MPB, for example with the normalized difference vegetation index (NDVI), a ratio 100 of red and NIR reflectance bands. Satellite imagery-based NDVI generally correlates well with 101 chl-a concentrations (Méléder et al., 2003; Barillé et al., 2011; Brito et al., 2013; Daggers et al., 102 2018; Jacobs et al., 2021), although the functional structures of calibration models vary 103 depending on sampling and analysis methodologies and the exact spectral wavelengths used for 104 NDVI calculations. Alternatively, mapping MPB with UAV-acquired multispectral data has 105 generally been limited to machine learning-based habitat classification rather than biomass 106 quantification (e.g., Román et al., 2023; Davies et al., 2023). As such, calibration models linking 107 UAV-generated NDVI or other spectral indices to chl-a, as well as methodologies to build such 108 models, are currently lacking. 109 Here, we develop and report on a protocol for linking data collected with the MicaSense 110 RedEdge-MX multispectral dual sensor with sediment MPB biofilm biomass in an intertidal 111 mudflat in British Columbia, Canada. The MicaSense RedEdge-MX is a commonly used UAV-112 mounted multispectral sensor, yet no standardized methodology currently exists for calibrating 113 spectral indices calculated from this type of senor with sediment photopigments for MPB 114 biomass quantification. Specifically, our aims were to (1) build an adaptable protocol for the 115 collection, processing, analysis, and calibration of UAV-acquired MicaSense RedEdge-MX 116 imagery and sediment photopigment field data, and (2) use the protocol to test a variety of 117 spectral indices and photopigment measurements to calculate a calibration equation specifically 118 for mapping MPB biomass in the Fraser River Estuary. 119 2. Materials and methods 120 2.1. Field data collection 121 Data used to build the calibration models consisted of UAV-acquired multispectral imagery and 122 georeferenced surface sediment samples, collected from three 4-27 ha sites within Roberts Bank, 123 Fraser River Estuary, British Columbia in March, 2023 (Table 1; Figure 1). Concurrent point-124 sampling and UAV surveys allowed us to pair photopigment measurements with exact pixels 125 from the multispectral imagery. 126 127 Multispectral imagery was collected with a DJI Matrice 300 RTK system equipped with a 128 MicaSense RedEdge-MX multispectral dual sensor (Table 2). This camera system can capture 10 129 spectral bands from 444-842 nm (Table 3). A downwelling light sensor (DLS) mounted to the 130 UAV and a ground-based calibrated reflectance panel were used to calibrate multispectral data 131 based on lighting conditions. Before flights, square polyethylene ground control points (GCP) 132 were placed on the study area to identify sediment sampling locations. 133 Using the GCPs as reference locations, we collected between 11-14 sediment samples 134 within one hour of UAV-acquisitions from each of the three survey sites, for a total of 38 135 sediment samples. We sampled the top 2 mm of sediment using a putty knife and a plastic 136 laminate sheet with a 10 x 10 cm square opening as a guide. Each sample was placed into a 120 137 mL polypropylene container and stored in the dark on dry ice in the field, maintained at -20 °C 138 until processing which occurred within 24 hours. 139 To evaluate how reflectance may varies across broad habitat types, we arbitrarily selected 140 a few representative locations at five basic habitat types in our study site (bare sediment, 141 intertidal biofilm, gravel, saltmarsh, and shallow water). 142 Table 1. Summary of unoccupied aerial vehicle (UAV) flight altitude above ground level (AGL), acquisition date and time, survey site area, and number of images collected, and number of sediment samples collected from each site. Site Altitude (m AGL) Overlap (%) Acquisition date Acquisition time Survey area (ha) Number of samples 1 50 90/75 10/03/2023 15:49 – 16:11 27 13 2 30 90/75 11/03/2023 14:36 – 14:59 7 14 3 30 90/75 11/03/2022 15:22 – 15:37 4 11 143 144 Figure 1. Collection of georeferenced sediment samples: (a) map of NDVI and surface scrape sediment sample locations from one of three survey sites; (b) 10 x 10 x 0.02 cm sediment sample (c) georeferencing scheme per location; (d) NDVI orthomosaic with sample location georeferencing guides overlayed. 145 (c) (d) (a) (b) B Table 2. Sensor specifications for Micasense RedEdge-MX Dual Camera System Pixel size 3.75 μm Resolution 1280 x 960 (1.2 MP x 5 imagers) Aspect ratio 4:3 Sensor size 4.8 mm x 3.6 mm Focal length 5.4 mm Field of view 47.2 degrees horizontal, 35.4 degrees vertical Output bit depth 12-bit GSD @ 120 m 8 cm/pixel per band 146 Table 3. Center wavelengths and bandwidths for Micasense RedEdge-MX Dual Camera System. Band number Band name Centre wavelength (nm) Bandwidth B1 Blue-444 444 28 B2 Blue 475 14 B3 Green-531 531 14 B4 Green 560 27 B5 Red-650 650 16 B6 Red 668 14 B7 Red edge-705 705 10 B8 Red edge 717 12 B9 Red edge-740 740 18 B10 NIR 842 57 147 2.2. Photogrammetric processing 148 A structure from motion photogrammetry methodology was implemented to produce 149 orthomosaics from the UAV-acquired 10-band multispectral imagery, with 3D point clouds and 150 digital surface models (DSMs) as intermediary products. First, sparse tie-point clouds for image 151 alignment were generated, and then dense point clouds developed by pixel-based depth map 152 generation, which were filtered moderately to remove unwanted noise and artifacts. From the 153 dense point cloud, we generated DSMs and 10-band orthomosaics with pixel sizes ranging from 154 0.024 to 0.051 m. The orthomosaics were radiometrically calibrated with images of the 155 calibration panel and data from the camera’s DLS, then atmospherically corrected using a dark 156 object subtraction algorithm (Chavez, 1996). We then calculated two versions of NDVI using 157 different wavelength combinations, green NDVI, and the Phytobenthos Index (PI), which has 158 been proposed to overcome limitations of NDVI for MPB quantification using satellite- or field 159 spectrophotometer-acquired multi- or hyperspectral reflectance data (Table 4). 160 161 Table 4. Multispectral single-band vegetation indices used to estimate microphytobenthos (MPB) biomass and pigment composition. Vegetation Index Abbr. Calculation MicaSense equivalent Reference Normalized Difference Vegetation Index with red-edge NDVI740 (R750 − R675) / (R750 + R675) (R740 − R668) / (R740 + R668) Jacobs et al., 2021 Normalized Difference Vegetation Index with NIR NDVI842 (R865 − R655) / (R865 + R655) (R842 − R668) / (R842 + R668) Daggers et al., 2018 Green NDVI GNDVI (R750 − R550) / (R750 + R550) (R740 − R560) / (R740 + R560) Rouse et al., 1973; Phytobenthos Index PI (R750 − R635) / (R750 + R635) (R740 − R650) / (R740 + R650) Barillé et al., 2011 162 2.3. Sediment photopigment quantification 163 Sediment samples for pigment analysis were thawed overnight and homogenized, then 1-2 g was 164 aliquoted to 10mL of refrigerated 90 % acetone in a 15-mL polypropylene centrifuge tube, which 165 were incubated for 24 h in the dark at 2° C. After incubation, the samples were centrifuged at 166 1500 rpm for 5 min. Acetone supernatant containing extracted photosynthetic pigments was 167 decanted into a clean 13 x 100 mm borosilicate culture tube. Concentrations of photosynthetic 168 pigments (chl-a and phæopigments) were then measured spectrofluorometrically according to 169 Arar & Collins (1997). Sediment pellets were re-weighted after ≥ 4 days of drying, then chl-a 170 and phæopigments contents (µg g-1) were calculated using Equation 1 and 2. Photopigment 171 contents were converted to concentrations (mg m-2) using the known area per surface scrape and 172 the total weight of each sample before subsampling. 173 1. 𝐶ℎ𝑙 𝑎 = 1.75472 ∗ (𝐹𝑜 − 𝐹𝑎) ∗ 1.3876 ∗ ( 𝑣 𝑔 ) 174 2. Phæo = 1.75472 ∗ ((2.325 ∗ Fa) − 𝐹𝑜 ∗ 1.3876 ∗ ( 𝑣 𝑔 ) 175 where Fo is the fluorescence reading before acidification (blank corrected), Fa is the fluorescence 176 reading after acidification (blank corrected), v is the volume of acetone extract (in litres), g is the 177 dry weight of sediment (in grams), and 1.75472 is an acid ratio correction factor. 178 179 2.4. Spectral reflectance to photopigment calibration models 180 To relate spectral reflectance to sediment photopigment content, we first calculated mean 181 spectral index values per sediment sample station located the UAV-acquired data. The exact area 182 and location of each surface scrape was delineated using the known dimension of the surface 183 scrapes and their distance and direction from the GCPs (Figure 1). Linear regression was used to 184 assess correlations between each of the 14 spectral indices and chl-a (chl-a with acidification to 185 remove phæopigments) or total pigments (chl-a + phæopigments) measured as either content (µg 186 g-1) or concentration (mg m-2). Significance level of α = 0.05 was set for all statistical analyses. 187 188 2.5. Software 189 Photogrammetry was performed AgiSoft Metashape Professional Version 2.1 (Agisoft LLC, St. 190 Petersburg, Russia). Creation polyline features for identification of surface scrapes on 191 orthomosaics and export of spectral index values was done in ArcGIS Pro 2.7.0 software (Esri, 192 USA). Statistical analysis was conducted using the statistical program R Studio version 193 2022.12.0 (cran.r.project.org). 194 3. Results 195 3.1. Spectral signatures of intertidal substrates 196 Analysis of the reflectance spectra from the 10-band imagery show that MPB biofilm has a 197 downwelling PAR absorption peak at 668 nm, a steep slope in reflectance from 668 to 705 nm, 198 and emittance peak at 717 nm which decreases slightly to 842 nm (Figure 2). This same pattern 199 was seen for the salt marsh, but with a relatively higher emittance peak at 717nm and a slight 200 increase from 740 to 842 nm. For bare mud, absorption and emittance peaks were much less 201 pronounced, whereas gravel and shallow water had lower reflectance in the red-edge and NIR 202 than the red bands. All substrates had the highest absorption in either the blue 444 or 475 nm 203 bands. 204 Figure 2. Maps of spectral indices and reflectance spectra of intertidal mudflat surfaces: (a) red-green- blue (RGB), (b) false-colour, (c) Normalized Difference Vegetation Index with near-infrared 842 nm (NDVI842), (d) Phytobenthos Index (PI). Panel (f) show a graph of reflectance spectra recorded by the MicaSense RedEdge-MX multispectral sensor. Signatures are mean spectral responses with offset and (e) dashed lines are standard deviation. Each spectrum corresponds to a location on the images identified by coloured circles representing various following surface classes. 205 3.2. Photosynthetic pigments 206 For all sites, mean chl-a content was 73 ± 34µg g-1 with a range of 27-145 µg g-1, and mean 207 phæopigment content was 18 ± 22 µg g-1 with a range of 2-71 µg g-1 (Table 5). Mean chl-a 208 concentration was 95 ± 22 mg m-2, ranging from 28-209 mg m-2, and mean phæopigment content 209 was 22 ± 27 mg m-2 with a range of 2-90 mg m-2. Chlorophyll-a to phæopigment mean value was 210 13 ± 10, with a range of 1-48. 211 Table 5. Results of photopigment chemical analyses and zonal statistics output of referenced sample sites from an NDVI orthomosaic. Name Site chl-a (µg g-1) phæo (µg g-1) chl-a (mg m-2) phæo (mg m-2) chl-a:phæo A1R 3 98.1564 46.2113 79.5955 37.4730 2.1241 A5R 3 97.8508 52.2446 115.5903 61.7161 1.8729 A6R 3 78.9704 46.2270 65.9861 38.6263 1.7083 A7R 3 141.2893 61.5197 179.6557 78.2251 2.2967 A7T 3 113.2744 56.2222 106.2409 52.7312 2.0148 A8R 3 43.2042 39.4313 55.7154 50.8499 1.0957 A8alt 3 121.6787 42.2507 125.5669 43.6008 2.8799 A9R 3 145.2056 70.6549 184.8172 89.9293 2.0551 A10T 3 53.7865 38.1481 99.6791 70.6974 1.4099 A10R 3 109.7875 56.6764 144.3139 74.5002 1.9371 A11T 3 73.7641 46.0122 110.8156 69.1239 1.6031 A12R 3 54.1688 1.9119 108.3439 3.8239 28.3331 A13R 3 39.9385 2.3421 39.5342 2.3184 17.0526 A4R 3 36.1693 2.4520 29.1827 1.9783 14.7511 BWR 1 27.0917 3.3731 34.0567 4.2403 8.0316 IR 1 34.6408 3.3872 35.1322 3.4352 10.2271 IIR 1 42.5446 3.0535 44.5082 3.1944 13.9333 IIL 1 42.5942 2.9544 34.1805 2.3708 14.4173 IIup 1 47.1488 2.4364 43.9717 2.2722 19.3517 IIIR 1 59.9853 2.6307 63.9391 2.8041 22.8016 IIITL 1 35.5835 3.0582 39.1346 3.3634 11.6354 IVR3 1 41.4018 3.3132 42.8907 3.4323 12.4962 VIR3 1 53.9681 2.8661 68.8140 3.6545 18.8297 VL3 1 56.7495 2.9893 80.4438 4.2374 18.9841 VL6 1 37.8283 2.3576 28.3575 1.7674 16.0450 VL12 1 117.7493 2.4483 209.1977 4.3497 48.0945 LOGR12 1 139.0876 5.8446 145.6893 6.1220 23.7977 BWT 2 55.2808 4.0810 69.0110 5.0946 13.5458 B2R 2 90.2063 8.8902 95.6745 9.4291 10.1467 B3R 2 85.4499 7.3352 102.1824 8.7716 11.6492 B4R 2 52.2687 4.3857 94.0778 7.8938 11.9179 B7R 2 68.3114 3.2215 135.8907 6.4084 21.2052 B8R 2 86.7430 3.7980 164.6302 7.2082 22.8392 B10R 2 90.2913 3.4220 169.0033 6.4051 26.3856 B11R 2 88.1863 3.0536 197.1724 6.8275 28.8791 B12R 2 98.1564 46.2113 79.5955 37.4730 2.1241 B13R 2 97.8508 52.2446 115.5903 61.7161 1.8729 B13R6 2 78.9704 46.2270 65.9861 38.6263 1.7083 212 3.2. Calibration model comparisons 213 Statistically significant correlations between spectral index values and photopigment 214 measurements were found for 11 of the 16 models evaluated, with 6 having an R2 > 0.6 (Figure 215 3; Table 5). Photopigment concentration (mg m-2) was generally a better predictor of spectral 216 index value than content (µg g-1). Among all spectral indices, NDVI842 correlated best with chl-a 217 concentration (R2 = 0.09), followed by NDVI740 with chl-a + phæopigment (R2 = 0.84). The 218 model equation was Chl 𝑎 = 744.38 ∗ NDVI842 + 17.913 (Fig. 3). 219 Figure 3. Relationships between sediment photopigments and spectral reflectance indices for the quantification of microphytobenthos in intertidal sediments. Chl-a = chlorophyll-a (chl-a); NDVI (740/842) = Normalized Difference Vegetation Index with red-edge 740/842 nm; (NPCI = Normalized Pigment Chlorophyll-a Index; PI = Phytobenthos Index. See Table 4 for wavelengths used. Top eight panels depict relationships with variables with weight per gram of dry sediment (µg g-1), and bottom eight panels depict relationships with variables per unit area (mg m-2). Turquoise and red trendlines show statistically significant and insignificant relationships, respectively. 220 221 222 223 224 Table 4. Linear regressions of sediment photopigments versus spectral reflectance indices. Chl-a = chlorophyll-a (chl-a); NDVI (740/842) = Normalized Difference Vegetation Index with red-edge 740/842 nm; NPCI = Normalized Pigment Chlorophyll-a Index; PI = Phytobenthos Index. Spectral index values have been standardized to a mean of 0 and a standard deviation of 1. Bold text shows statistically significant relationships. See Table 4 for wavelengths used. chl-a (µg g-1) chl-a + phæopigments (µg g-1) intercept slope R2 p intercept slope R2 p NDVI (740) 71.0 26.5 0.62 2.9 x 10-9 90.2 33.9 0.46 1.6 x 10-6 NDVI (842) 71.0 27.0 0.64 9.9 x 10-10 90.2 34.8 0.49 6.3 x 10-7 GNDVI 71.0 5.6 0.0014 0.31 90.2 -6.4 -0.01 0.43 PI 71.0 11.4 0.091 0.036 90.2 2.7 -0.024 0.74 chl-a (mg m-2) chl-a + phæopigments (mg m-2) intercept slope R2 p intercept slope R2 p NDVI (740) 92.0 45.9 0.76 7.1 x 10-13 115.3 55.7 0.69 8.0 x 10-11 NDVI (842) 92.0 50.0 0.90 2.2 x 10-16 115.3 61.3 0.84 5.2 x 10-16 GNDVI 92.0 23.8 0.18 4.2 x 10-3 115.3 10.2 -0.0039 0.36 PI 92.0 30.7 0.32 1.1 x 10-4 115.3 21.56 0.076 0.052 225 4. Discussion and perspectives 226 We developed a protocol for calibrating MicaSense RedEdge-MX dual sensor multispectral data 227 with measurements of sediment photopigments that provided promising results. A correlation 228 was found between NDVI842 and chl-a concentration that explained 90 % of the variation, with 229 the R2 of five other models exceeding 0.6. This high predictive power was achieved by designing 230 our protocol to link spectral index values and photopigment concentrations as closely as possible. 231 First, sediment samples for photopigment analysis were collected following the UAV surveys, so 232 that the mudflat surface at each sample site remained undisturbed in the imagery. Secondly, by 233 georeferencing the samples with GCPs, we were able to pair the 10 x 10 cm surface scrapes with 234 the exact 4-6 corresponding 2 x 2 cm GSD pixels of the sample site from a multispectral imagery 235 collected a low elevation above ground level (30 – 50 m). Due to this close link between imagery 236 and sediment samples, the calibration models developed here outperformed many of those 237 reported for satellite-based spectral indices, whereby photopigment measurements from one or 238 several sediment point samples were used to represent pixels ≥ 10 m. 239 Building a valid spectral index to MPB biomass calibration model requires the collection 240 of sediment samples containing a wide range of photopigment concentrations representative of 241 those found in the estuary under investigation. Furthermore, the time between image acquisition 242 and sediment collection should be minimized to reduce the potential effect of diel changes to 243 surficial photopigment concentrations from vertical MPB migration. The biogeomorphology of 244 the site allowed the collection of a range of sediments with minimal logistical challenges and 245 time expenditure. Roberts Bank is characterized by mudflats with very shallow hummocks 246 colonized by diatomaceous MPB biofilm (relatively low chl-a) from the edge of the saltmarsh to 247 ~100 m offshore, beyond which it transitions into a ridge and runnel complex that is colonized 248 by cyanobacterial-dominant biomat (high chl-a). By travelling only ~200 m offshore, we were 249 able to collect sediment samples containing 28-209 mg m-2 chl-a, and the ~400 m round-trip 250 allowed for sediment sample collection to occur within one hour of UAV surveys. Additionally, 251 identifying sample sites with highly visible GCPs further minimized the time expended on 252 flagging sites before UAV flight and returning to them for sediment collection after UAV 253 surveys; thus, GCPs are preferred over recording location information with handheld GPS or 254 RTK units. 255 Despite the wide range of sediment chl-a and total pigment concentrations found here, we 256 observed no asymptotic behaviour of spectral indices at chl-a > 100 mg m-2 as had been reported 257 in several previous studies (Méléder et al., 2003, 2010; Barillé et al., 2011; Brito et al., 2013). 258 Our methodological approach differed substantially from most of those that reported saturation; 259 for example, Méléder et al. (2003) and Barillé et al. (2011) used monospecific cultures of 260 diatoms and quantified photopigment by high performance liquid chromatography (HPLC) as 261 opposed to fluorescence. Sediment-based studies involve complex compositions of substrates, 262 organic matter, and water content, allowing for a more diverse range of reflectance values and a 263 potentially broader range of NDVI-chl-a relationships. Our results were more similar to those 264 from studies that measured photopigments from sediment with fluorometry or spectroscopy, 265 especially those that used in situ field reflectance measurements (Daggers et al., 2018). 266 Among the photopigment measurements tested, we found that chl-a concentration, as 267 opposed to content, correlated best with spectral index value. Concentration is generally regarded 268 as preferable to content for measurements of sediment biogeochemical properties, as content 269 measurements are confounded by various factors (e.g., core density per mass; Tolhurst et al., 270 2005). However, we used content as an intermediate step to calculate areal concentration rather 271 than directly measuring sediment volume, so our concentration measurements were subject to the 272 same confounding factors. By strict definition, mg m-2 is a measurement of surface density but is 273 referred to as concentration by convention in remote sensing and estuarine science literature. 274 Calculations of concentration vary by study, with some using volume and dry bulk density (e.g., 275 Daggers et al., 2018) and others scaling content by the area and weight of the total sample, as we 276 did (e.g., Kromkamp et al., 2006). Modifying our calculation to use sediment volume may 277 further improve the calibration model. Nevertheless, photopigments reported as areal 278 concentration remains the standard for satellite remote sensing-based MPB biofilm research and 279 is the better unit of measurement for relating photopigments to UAV-acquired spectral 280 information. 281 Interestingly, relationships between spectral indices and chl-a concentration were 282 stronger than those of total pigment (chl-a and phæopigment). Calibration models in many 283 studies have been built from NDVI and total pigments, as NDVI is understood to be more 284 sensitive to chl-a and related decomposition pigments than to chl-a alone (Antoine & Morel 285 1996; Jacobs et al., 2021). We found that modelling NDVI842 vs total pigments decreased the 286 explanatory power by ~8 % compared to acidified chl-a. We posit that the high correlation of 287 NDVI842 from MicaSense RedEdge-MX dual sensor data and acidified chl-a concentration is 288 advantageous compared to total pigment concentration, and more practical for monitoring the 289 ecologically-important live fraction of MPB. 290 Previous studies often report high correlations between satellite imagery-based NDVI and 291 sediment photopigments (e.g., Méléder et al., 2003; Barillé et al., 2011; Daggers et al., 2018), but 292 only weakly or moderately in others (Murphy et al., 2005; Kwon et al., 2016). This may be due 293 in part to the methodological differences outlined above, as well as the specific sensors and 294 wavelengths used for NDVI calculations. Since band specifications vary slightly across sensors 295 satellites, there is no single definition of NDVI. Practically, however, sensor-specific NDVIs can 296 be easily calibrated against a common reference sensor or photopigment concentration, allowing 297 for multi-sensor monitoring with UAV- and satellite-acquired multispectral data. As such, our 298 protocol calibrating the MicaSense RedEdge-MX dual sensor with sediment photopigment 299 concentration will be a valuable asset for combining UAV- and satellite acquired images, such as 300 resolving issues of spectral mixing within pixels, improving extrapolations aimed at upscaling, 301 and creating hybrid data products. 302 Conclusion 303 Here, we developed a protocol for calibrating UAV-mounted MicaSense RedEdge-MX dual 304 sensor multispectral data to sediment MPB biomass, as measured by photopigment content. Our 305 results suggest that calibrated spectral index values from MicaSense RedEdge-MX data can 306 provide a very accurate measurement of MPB biomass, able to achieve 90 % correlation between 307 NDVI842 and chl-a concentration. The strength of the calibration model was achieved by closely 308 pairing sediment photopigment concentration from georeferenced surface scrapes to the exact 309 corresponding pixels of the sample site from a multispectral orthomosaic. This was done by (1) 310 choosing an accessible and logistically manageable study area, (2) ensuring the area had a wide 311 range of sediment photopigment content, and (3) collecting sediment sample within one hour of 312 MicaSense imagery. Using this protocol, applications of UAV-acquired MicaSense RedEdge-313 MX multispectral imagery of mudflats can extend beyond habitat classification to elucidating 314 fine-scale spatial heterogeneity and short-term temporal dynamics of MPB biomass and bridging 315 the gap between UAV- and satellite- acquired imagery of intertidal ecosystems. 316 5. References 317 Antoine D & Morel A. (1996). Oceanic primary production: 1. Adaptation of a spectral light 318 photosynthesis model in view of application to satellite chlorophyll observations. Global 319 Biogeochem Cycles; 10: 43–55. 320 Arar EJ, Collins GB. In vitro determination of chlorophyll a and pheophytin a in marine and 321 freshwater algae by fluorescence. National Exposure Research Laboratory Report 1997; 322 445.0–1. 323 Barillé, L., Mouget, J., Méléder, V., Rosa, P., & Jesus, B. (2011). Spectral response of benthic 324 diatoms with different sediment backgrounds. Remote Sensing of Environment, 115(4), 325 1034-1042. https://doi.org/10.1016/j.rse.2010.12.008 326 Brito, A. C., Benyoucef, I., Jesus, B., Brotas, V., Gernez, P., Mendes, C. R., Launeau, P., Dias, 327 M. P., & Barillé, L. (2013). Seasonality of microphytobenthos revealed by remote-328 sensing in a south european estuary. Continental Shelf Research, 66, 83-91. 329 https://doi.org/10.1016/j.csr.2013.07.004 330 Brunier, G., Oiry, S., Lachaussée, N., Barillé, L., Le Fouest, V., & Méléder, V. (2022). A 331 machine-learning approach to intertidal mudflat mapping combining multispectral 332 reflectance and geomorphology from UAV-based monitoring. Remote Sensing (Basel, 333 Switzerland), 14(22), 5857. https://doi.org/10.3390/rs14225857 334 Chavez, J.P.S. (1996). Image-Based Atmospheric Corrections—Revisited and Improved. 335 Photogrammetric Engineering and Remote Sensing, 62, 1025-1036. 336 https://doi.org/10.1016/j.csr.2013.07.004 Daggers, T.D., Kromkamp, J.C., Herman, P.M.J., van der Wal, D. (2018). A model to assess 337 microphytobenthic primary production in tidal systems using satellite remote sensing. 338 Remote Sens. Environ.: 211, 129–145. 339 Davies, B. F. R., Gernez, P., Geraud, A., Oiry, S., Rosa, P., Zoffoli, M. L., & Barillé, L. (2023). 340 Multi- and hyperspectral classification of soft-bottom intertidal vegetation using a 341 spectral library for coastal biodiversity remote sensing. Remote Sensing of Environment, 342 290, 113554. https://doi.org/10.1016/j.rse.2023.113554 343 Douglas, T. J., Coops, N. C., & Drever, M. C. (2023). UAV-acquired imagery with 344 photogrammetry provides accurate measures of mudflat elevation gradients and 345 microtopography for investigating microphytobenthos patterning. Science of Remote 346 Sensing, 7, 100089. https://doi.org/10.1016/j.srs.2023.100089 347 Hope, J. A., Paterson, D. M., Thrush, S. F., & Van Alstyne, K. (2020). The role of 348 microphytobenthos in soft‐sediment ecological networks and their contribution to the 349 delivery of multiple ecosystem services. The Journal of Ecology, 108(3), 815-830. 350 https://doi.org/10.1111/1365-2745.13322 351 Jacobs, P., Pitarch, J., Kromkamp, J. C., & Philippart, C. J. M. (2021). Assessing biomass and 352 primary production of microphytobenthos in depositional coastal systems using spectral 353 information. PloS One, 16(7), e0246012. https://doi.org/10.1371/journal.pone.0246012 354 Kromkamp, JC, Morris, EP, Forster, RM, Honeywill C, Hagerthey, S, Paterson, DM. (2006) 355 Relationship of intertidal surface sediment chlorophyll concentration to hyperspectral 356 reflectance and chlorophyll fluorescence. Estuaries and coasts 29 (2), 183-196 357 Kwon BO, Lee Y, Park J, Ryu J, Hong S, Son S, et al. (2016) Temporal dynamics and spatial 358 heterogeneity of microalgal biomass in recently reclaimed intertidal flats of the 359 Saemangeum area, Korea. J Sea Res 116: 1–11. 360 MacIntyre, H. L., Geider, R. J., and Miller, D. C. (1996). Microphytobenthos: the ecological role 361 of the “secret garden” of unvegetated, shallow-water marine habitats. I. Distribution, 362 abundance and primary production. Estuaries 19(2A):186-201 [doi:10.2307/1352224]. 363 Méléder, V., Barillé, L., Launeau, P., Carrère, V., & Rincé, Y. (2003). Spectrometric constraint 364 in analysis of benthic diatom biomass using monospecific cultures. Remote Sensing of 365 Environment, 88(4), 386-400. https://doi.org/10.1016/j.rse.2003.08.009 366 Meleder, V.; Launeau, P.; Barille, L.; Combe, J. P.; Carrere, V.; Jesus, B.; Verpoorter, C. (2010). 367 Hyperspectral imaging for mapping microphytobenthos in coastal areas. In Geomatic 368 solutions for coastal environments; Maanan, M.; Robin, M., Eds.; Nova Science 369 Publishers, Inc.; pp. 71–139. 370 Murphy, RJ, Tolhurst, MG, Chapman, MG, Underwood, AJ. (2005) Estimation of surface 371 chlorophyll-a on an emersed mudflat using field spectrometry: accuracy of ratios and 372 derivative-based approaches. Int. J. Remote Sens., 26 (9), 1835-1859. 373 Oiry, S., & Barillé, L. (2021). Using sentinel-2 satellite imagery to develop microphytobenthos-374 based water quality indices in estuaries. Ecological Indicators, 121, 107184. 375 https://doi.org/10.1016/j.ecolind.2020.107184 376 Pinckney, J. L., & Lee, A. R. (2008). Spatiotemporal patterns of subtidal benthic microalgal 377 biomass and community composition in galveston bay, texas, USA. Estuaries and Coasts, 378 31(2), 444-454. https://doi.org/10.1007/s12237-007-9020-9 379 https://doi.org/10.1016/j.rse.2003.08.009 https://doi.org/10.1007/s12237-007-9020-9 Román, A., Prasyad, H., Oiry, S., Davies, B. F. R., Brunier, G., & Barillé, L. (2023). Mapping 380 intertidal oyster farms using unmanned aerial vehicles (UAV) high-resolution 381 multispectral data. Estuarine, Coastal and Shelf Science, 291, 108432. 382 https://doi.org/10.1016/j.ecss.2023.108432 383 Rouse, JW, Haas, RH, Schell, JA, Deering, DW. (1973) Monitoring vegetation systems in the 384 Great Plains with ERTS. Proceedings of the Third ERTS Symposium, NASA, SP-351, 385 vol.1 (Washington, DC: NASA), 309–317. 386 Tolhurst, T. J., Underwood, A. J., Perkins, R. G., & Chapman, M. G. (2005). Content versus 387 concentration; effects of units on measuring the biogeochemical properties of soft 388 sediments. Estuarine, Coastal and Shelf Science, 63(4), 665-673. 389 https://doi.org/10.1016/j.ecss.2005.01.010 390 391 392 https://doi.org/10.1016/j.ecss.2005.01.010 Supplementary material 393 394 Table S1. Sensor specifications for Micasense RedEdge-MX Dual Camera System Pixel size 3.75 μm Resolution 1280 x 960 (1.2 MP x 5 imagers) Aspect ratio 4:3 Sensor size 4.8 mm x 3.6 mm Focal length 5.4 mm Field of view 47.2 degrees horizontal, 35.4 degrees vertical Output bit depth 12-bit GSD @ 120 m 8 cm/pixel per band 395 S2. Photosynthetic pigments 396 For all sites, mean chl-a content was 73 ± 34µg g-1 with a range of 27-145 µg g-1, and mean 397 phæopigment content was 18 ± 22 µg g-1 with a range of 2-71 µg g-1 (Table 5). Mean chl-a 398 concentration was 95 ± 22 mg m-2, ranging from 28-209 mg m-2, and mean phæopigment content 399 was 22 ± 27 mg m-2 with a range of 2-90 mg m-2. Chlorophyll-a to phæopigment mean value was 400 13 ± 10, with a range of 1-48. 401 402 Table S3. Results of photopigment chemical analyses and zonal statistics output of referenced sample sites from an NDVI orthomosaic. Name Site chl-a (µg g-1) phæo (µg g-1) chl-a (mg m-2) phæo (mg m-2) chl-a:phæo A1R 3 98.1564 46.2113 79.5955 37.4730 2.1241 A5R 3 97.8508 52.2446 115.5903 61.7161 1.8729 A6R 3 78.9704 46.2270 65.9861 38.6263 1.7083 A7R 3 141.2893 61.5197 179.6557 78.2251 2.2967 A7T 3 113.2744 56.2222 106.2409 52.7312 2.0148 A8R 3 43.2042 39.4313 55.7154 50.8499 1.0957 A8alt 3 121.6787 42.2507 125.5669 43.6008 2.8799 A9R 3 145.2056 70.6549 184.8172 89.9293 2.0551 A10T 3 53.7865 38.1481 99.6791 70.6974 1.4099 A10R 3 109.7875 56.6764 144.3139 74.5002 1.9371 A11T 3 73.7641 46.0122 110.8156 69.1239 1.6031 A12R 3 54.1688 1.9119 108.3439 3.8239 28.3331 A13R 3 39.9385 2.3421 39.5342 2.3184 17.0526 A4R 3 36.1693 2.4520 29.1827 1.9783 14.7511 BWR 1 27.0917 3.3731 34.0567 4.2403 8.0316 IR 1 34.6408 3.3872 35.1322 3.4352 10.2271 IIR 1 42.5446 3.0535 44.5082 3.1944 13.9333 IIL 1 42.5942 2.9544 34.1805 2.3708 14.4173 IIup 1 47.1488 2.4364 43.9717 2.2722 19.3517 IIIR 1 59.9853 2.6307 63.9391 2.8041 22.8016 IIITL 1 35.5835 3.0582 39.1346 3.3634 11.6354 IVR3 1 41.4018 3.3132 42.8907 3.4323 12.4962 VIR3 1 53.9681 2.8661 68.8140 3.6545 18.8297 VL3 1 56.7495 2.9893 80.4438 4.2374 18.9841 VL6 1 37.8283 2.3576 28.3575 1.7674 16.0450 VL12 1 117.7493 2.4483 209.1977 4.3497 48.0945 LOGR12 1 139.0876 5.8446 145.6893 6.1220 23.7977 BWT 2 55.2808 4.0810 69.0110 5.0946 13.5458 B2R 2 90.2063 8.8902 95.6745 9.4291 10.1467 B3R 2 85.4499 7.3352 102.1824 8.7716 11.6492 B4R 2 52.2687 4.3857 94.0778 7.8938 11.9179 B7R 2 68.3114 3.2215 135.8907 6.4084 21.2052 B8R 2 86.7430 3.7980 164.6302 7.2082 22.8392 B10R 2 90.2913 3.4220 169.0033 6.4051 26.3856 B11R 2 88.1863 3.0536 197.1724 6.8275 28.8791 B12R 2 98.1564 46.2113 79.5955 37.4730 2.1241 B13R 2 97.8508 52.2446 115.5903 61.7161 1.8729 B13R6 2 78.9704 46.2270 65.9861 38.6263 1.7083 403 404 Abstract 1. Introduction 2. Materials and methods 2.1. Field data collection 2.2. Photogrammetric processing 2.3. Sediment photopigment quantification 2.4. Spectral reflectance to photopigment calibration models 2.5. Software 3. Results 3.1. Spectral signatures of intertidal substrates 3.2. Photosynthetic pigments 3.2. Calibration model comparisons 4. Discussion and perspectives Conclusion 5. References S2. Photosynthetic pigments