Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 Contents lists available at ScienceDirect Int J Appl Earth Obs Geoinformation journal homepage: www.elsevier.com/locate/jag Comparison of four UAV georeferencing methods for environmental T monitoring purposes focusing on the combined use with airborne and satellite remote sensing platforms ⁎ Joan-Cristian Padróa, , Francisco-Javier Muñozb, Jordi Planasc, Xavier Ponsd a Grumets Research Group, Departament de Geografia, Universitat Autònoma de Barcelona, Departament de Geografia Office B1092, Edifici B. Universitat Autònoma de Barcelona, 08193 Bellaterra, Catalonia, Spain b HEMAV S.L. Edifici RDIT. Office 0007. C\Esteve Terrades, 1. Parc Mediterrani de la Tecnologia, 08860 Castelldefels, Catalonia, Spain c Universitat Autònoma de Barcelona, Departament de Geografia Office B1092. Edifici B, Universitat Autònoma de Barcelona, 08193 Bellaterra, Catalonia, Spain d Grumets Research Group, Departament de Geografia, Universitat Autònoma de Barcelona, Departament de Geografia Office B1094. Edifici B. Universitat Autònoma de Barcelona. 08193 Bellaterra, Catalonia, Spain A R T I C LE I N FO A B S T R A C T Keywords: This work is aimed at the environmental remote sensing community that uses UAV optical frame imagery in UAV combination with airborne and satellite data. Taking into account the economic costs involved and the time Drone investment, we evaluated the fit-for-purpose accuracy of four positioning methods of UAV-acquired imagery: 1) Direct georeferencing direct georeferencing using the onboard raw GNSS (GNSSNAV) data, 2) direct georeferencing using Post- Indirect georeferencing Processed Kinematic single-frequency carrier-phase without in situ ground support (PPK1), 3) direct georefer- ASPRS standards encing using Post-Processed Kinematic double-frequency carrier-phase GNSS data with in situ ground support Post-Processed Kinematic (PPK2), and 4) indirect georeferencing using Ground Control Points (GCP). We tested a multispectral sensor and an RGB sensor, onboard multicopter platforms. Orthophotomosaics at < 0.05 m spatial resolution were gener- ated with photogrammetric software. The UAV image absolute accuracy was evaluated according to the ASPRS standards, wherein we used a set of GCPs as reference coordinates, which we surveyed with a differential GNSS static receiver. The raw onboard GNSSNAV solution yielded horizontal (radial) accuracies of RMSEr≤1.062 m and vertical accuracies of RMSEz≤4.209 m; PPK1 solution gave decimetric accuracies of RMSEr≤0.256 m and RMSEz≤0.238 m; PPK2 solution, gave centimetric accuracies of RMSEr≤0.036 m and RMSEz≤0.036 m. These results were further improved by using the GCP solution, which yielded accuracies of RMSEr≤0.023 m and RMSEz≤0.030 m. GNSSNAV solution is a fast and low-cost option that is useful for UAV imagery in combination with remote sensing products, such as Sentinel-2 satellite data. PPK1, which can register UAV imagery with remote sensing products up to 0.25 m pixel size, as WorldView-like satellite imagery, airborne lidar or orthoi- magery, has a higher economic cost than the GNSSNAV solution. PPK2 is an acceptable option for registering remote sensing products of up to 0.05 m pixel size, as with other UAV images. Moreover, PPK2 can obtain accuracies that are approximate to the usual UAV pixel size (e.g. co-register in multitemporal studies), but it is more expensive than PPK1. Although indirect georeferencing can obtain the highest accuracy, it is nevertheless a time-consuming task, particularly if many GCPs have to be placed. The paper also provides the approximate cost of each solution. Abbreviations: APC, Antenna Phase Centre; ASPRS, American Society for Photogrammetry and Remote Sensing; AT, Aerial-Triangulation; DG, Direct Georeferencing; DGNSS, Differential Global Navigation Satellite System; EOPs, Exterior Orientation Parameters; ETRS89, European Terrestrial Reference System 1989; DSLR, Digital Single-Lens Reflex; GCP, Ground Control Points; GLONASS, Global'naya Navigatsionnaya Sputnikovaya Sistema; GNSS, Global Navigation Satellite System; GNSSNAV, Navigation GNSS onboard the UAV; GPS, Navstar Global Positioning System; ICP, Independent Check Points; IG, Indirect Georeferencing; IMU, Inertial Measurement Unit; ISO, Integrated Sensor Orientation; NSSDA, National Standard for Spatial Data Accuracy; NVA, NonVegetated Area; PC, Perspective Centre; PP, PostProcessed; PPK, PostProcessed Kinematic; PRS, Permanent Reference Station; RMSE, Root Mean Square Error; RTK, RealTime Kinematic; SRS, patial resolution; SfM, Structure from Motion; UAV, Unmanned Aerial Vehicle; VBS, Virtual Base Station ⁎ Corresponding author. E-mail addresses: JoanCristian.Padro@uab.cat (J.-C. Padró), jmunoz@hemav.com (F.-J. Muñoz), j.planas@creaf.uab.cat (J. Planas), Xavier.Pons@uab.cat (X. Pons). https://doi.org/10.1016/j.jag.2018.10.018 Received 4 July 2018; Received in revised form 2 October 2018; Accepted 25 October 2018 Available online 06 November 2018 0303-2434/ © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/). J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 1. Introduction et al., 2004). In environmental and topographic mapping projects, usually there is no requirement for Real-Time Kinematic (RTK) geor- Capturing images by means of optical sensors installed in < 25 kg eferencing. Nevertheless, it is important to obtain an imagery posi- Unmanned Aerial Vehicles (UAV) or drones is gaining importance in tioning that is as accurate as possible. This can be realized by post- quantitative remote sensing (Aasen et al., 2018), environmental re- processing the sensor central coordinates at the image exposure time, search (Manfreda et al., 2018), remote sensing applications for vege- which takes into account the UAV trajectory, sensor position at trigger tated areas (Komárek et al., 2018), precision agriculture (Borgogno time (Rehak and Skaloud, 2017), satellite geometries and atmospheric Mondino and Gajetti, 2017), highly detailed metrology (Daakir et al., error sources. The UAV trajectory is determined by interpolating the 2017) and cartographic (Crommelinck et al., 2017) applications. onboard GNSS data, whereby the sensor position at trigger time re- Moreover, their use in combination with other environmental remote quires synchronizing between the camera and the onboard GNSS, and sensing products, such as high Spatial Resolution (SR) satellite imagery the GNSS satellite constellation information can be provided by Per- (Lin et al., 2011) or airborne orthophotomaps of official mapping manent Reference Stations (PRS) or Virtual Base Stations (VBS). This agencies (Whitehead and Hugenholtz, 2014) has attracted growing in- solution was tested in UAV platforms combining a single-frequency low terest. In many European countries, current legislation allows a UAV to cost GNSS receiver and IMU data (Turner et al., 2014a,b; Rehak et al., fly at a certain maximum height (e.g., 120 m in Spain (BOE, 2017). 2013), achieving accuracies almost similar to those of the typical spatial Consequently, commercial sensors usually provide SR between 0.01 m resolution of optical sensors suitable in a low-weight UAV for remote (1 cm) in RGB sensors and 0.10 m (10 cm) in multispectral sensors. A sensing applications. Other studies (Bláha et al., 2012; Rehak and problem of this ultra-high SR is that it usually exceeds the accuracy of Skaloud, 2015) analyzed the capacity of post-processing GNSS and IMU direct Exterior Orientation Parameters (EOPs) determination of in- sensor integration data in UAV to outperform DG providing results of dividual images at the sensor exposure time, i.e., the positional (X,Y,Z) centimetric accuracies. Chio (Chio, 2016) evaluated a VBS RTK GPS- accuracy directly obtained from the platform onboard Global Naviga- assisted self-calibration bundle adjustment for aerial triangulation of tion Satellite System (GNSS) receiver, and the sensor orientation UAV images for updating 1:5000 topographic maps; he collected the (ω,φ,κ) directly obtained from the Inertial Measurement Unit (IMU) data by flying a fixed-wing UAS mounted with a 24-mm focal-length device. This particular process of determining EOPs for individual Canon EOS 5D Mark II camera at 550 m, with 0.27 m in planimetry and images at the exposure time entails an inaccurate georeferencing of the 0.24 m in height accuracies (RMSE). In a later study, Chio and Lin resulting orthophotomosaic, or at least an error largely exceeding the (2017) tested a thermal camera mounted on a UAV using a Trimble pixel size. The receiver rate of position measurements per second (Hz) BD970GNSS and PPK data that gave good preliminary results. All these is crucial for sampling the position of a mobile object; for example, at a aforementioned precedents use GNSS and IMU post-processed data to velocity of 5 m/s, a 10 Hz GNSS receiver samples a position every 0.5 m. directly determine EOPs (X,Y,Z; ω,φ,κ) before the bundle adjustment On the other hand, a receiver can work in code or carrier-phase mode, phase. The determination of the image position still depends greatly on the latter usually provides more range accuracy, since the receiver is the GNSS receiver. Furthermore, the Integrated Sensor Orientation capable of distinguishing the sinusoidal wave oscillation where the (ISO), which is commonly referred to as the pose of the image (Aasen signal code is located. This solution can be obtained in an expedient et al., 2018; Wouda et al., 2016), is not a trivial procedure for a non- manner by using a single-frequency receiver. Moreover, the final ac- specialized user. curacy can be improved by using a dual-frequency differential GNSS In our study, we compare the geometric accuracy of four different (DGNSS) receiver. A DGNSS is capable of reading two carrier-phases at georeferencing techniques: (1) onboard Navigation GNSS (GNSSNAV); two wavelengths (typically L1/L2) as well as consider signal delays (2) PPK single frequency and the corrections of a permanent reference caused by atmospheric effects (differential measurement). As reported station located 4 km away from the flight area (PPK1); (3) PPK dual by Colomina and Molina (2014), the Direct Georeferencing (DG) ac- frequency and the corrections of a base station located in the flight area; curacy of a conventional UAV depends on the quality of the GNSS re- and (4) Ground Control Points and Independent Check Points ceiver; for example, in precision code, the positioning error can reach (GCP–ICP). The evaluation of GNSSNAV and PPK methods consisted of 0.77 m, and in the L1/L2 carrier-phase, the error can reach 0.01 m a comparison of image-derived coordinates with well-determined sur- (Colomina and Molina, 2014). The code-based moderate accuracy is veyed coordinates on specifically materialized targets used as ICP acceptable in UAV campaigns that only require low metric quality (RMSEx, RMSEy, RMSEr and RMSEz), while cross-validation was per- (Remondino et al., 2012). However, this imagery can neither be used to formed for the GCP–ICP method. generate detailed cartography at large scales, such as 1:200 (ASPRS, 2014), nor be assumed to have reasonable pixel co-registration in multi- 2. Materials and methods temporal studies using imagery with an SR < 0.1 m. A classic precise technique for accurate georeferencing of aerial imagery, known as In- 2.1. Study area direct Georeferencing (IG), consists of using Ground Control Points (GCPs) to support Aero-Triangulation (AT) and reaches accuracies of We used the test area of the Can Gelabert farm (municipality of 0.02 m, which was evaluated by using Independent Check Points (ICPs) Riner, Catalonia) located northeast of the Iberian Peninsula, in the (Turner et al., 2014a, 2014b). IG was used in (Tian et al., 2017) to Mediterranean basin between the Catalan Central Depression and the compare WorldView-2 imagery and UAV imagery. It provided pixel size Pre-Pyrenees. The planned study area has an area of 60 000 m2 (6 ha), accuracies of 1 cm RGB imagery and 7 cm multispectral (MicaSense forming a rectangle having a side of 300 m in the E–W direction and a RedEdge) imagery. Also (Zabala, 2017) used GCP to compare Sentinel-2 side of 200 m in the N–S direction. The altitude of the terrain at the and MicaSense RedEdge data. The drawback is that taking data on the materialized points is between 719.01 m and 734.20 m. The terrain ground is not always feasible. When it is, it is very costly and entails a morphology is not absolutely flat, as there are some unevenness formed high volume of time and resources, which is ultimately not optimal or by abandoned vineyards (including plain benches ranging from a worth the overall investment of effort. minimum slope of 0° to a maximum slope of near 90° when moving Post-Processing Kinematic (PPK) DG solution has been a widely from one bench to another) as well as trees higher than 8 m. In the used method in airborne platforms since 1990 (Skaloud et al., 1996). central zone, there is a field of cereal crops, and the woody vegetation Recently, the market offers PPK solutions with devices light enough to of the area is formed mainly by Quercus humilis, Quercus ilex, Ulmus be carried in < 25 kg UAV platforms. Consequently, PPK is an inter- minor and Pinus nigra ssp. salzmanii. The man-made elements consist of esting option for remote sensing environmental users due to the accu- an 18th century rural complex, which includes the house of Can rate processing procedure after the flight (not in real-time, (Bisnath Gelabert (floor plan of 325 m2), two haystacks, a pond and a small well. 131 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 Fig. 1. a) Gelabert farm (Riner, Solsonès County). Red square shows the scope of the planned flight area. Background: Official cartography (SR = 0.25 m) (ICGC, 2018). b) Location of the area NE of the Iberian Peninsula. c) Location of the area in the European context. d) Digital Elevation Model of the planned flight area (Inverse Distance Weighted interpolation from official lidar cartography (SR = 2 m) (ICGC, 2018) (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article). A rural area, such as the one selected, proved to be an exemplary area created, wherein four quadrants of 25 cm could be easily painted on the for land cover mapping, environmental monitoring as well as for ar- base, thereby forming a checkered layout. This further ensured the cheological purposes. Apart from this diversity, the test field followed necessary orthogonality for facilitating the location of the target center. the restrictions imposed by the Spanish aerial security agency (Agencia A DGNSS static receiver, which was supported on a tripod with a clamp Estatal de Seguridad Aérea, AESA) (BOE, 2017), which prohibits flying for high stability, was used to accurately determine the position of the UAVs in urban areas, in areas with large groups of people, or in the GCPs (Fig. 3). vicinity of aerodromes or airports (Fig. 1). The DGNSS receiver (Leica Geosystems, 2017) can work with the Eight flat concrete bases of 75 cm x 75 cm x 10 cm were purposely information provided by a network of Permanent Reference Station built within the study area in order to ensure the solidity and durability (PRS), but this infrastructure has to be implemented in the region of the of these points. These bases, which were used as GCPs and ICPs, were study area, as with the case of the European permanent reference sta- distributed in the study area (Martínez-Carricondo et al., 2018) and tion network (EUREF-EPN) (Ihde et al., 2014) or Continuously Oper- covered the northing and easting ranges of the flight area. The location ating Reference Stations (CORS) (Snay and Soler, 2008) in North of the bases was planned to focus on the constructed complex and the America. For example, we used information from the PRS located in main crop field surrounding the central flight area. The GCPs were all Solsona (SONA, http://catnet-ip.icc.cat/spiderweb/frmIndex.aspx), characterized by being located in plain areas not covered by vegetation. which is managed by the Institut Cartogràfic i Geològic de Catalunya However, horizon visibility was not the same in all GCPs, as the sky (ICGC, 2017) and located about 4 km from our study area. The DGNSS view factor was > 89% in all GCPs, except in GCP Nr 7 (77%) (Table 1). measurements combined with the information of PRS allow the re- These differences are important for the visibility of the satellites when ceiver’s position to be determined more precisely than with RTK mea- measuring the GCP coordinates (Fig. 2). surements. Nevertheless, these measurements require post-processing, A steel screw was placed at the center of each base to position the and thus more time for obtaining the final positions. Continuous ob- surveying rod used for determining the coordinate. A template was also servation of the reference stations makes it possible to calculate 132 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 Table 1 Sky view factor (SVF) in the surveyed points used as reference coordinates. Point-ID Point-1 Point-2 Point-3 Point-4 Point-5 Point-6 Point-7 Point-8 SVF (%) 94.93 96.04 97.15 94.23 89.42 95.90 77.07 98.92 differential corrections to improve the accuracy of the broadcast orbits, study has been widely used in previous studies to process UAV images the synchrony of the clocks of the satellites and receivers, and the and generate orthophotomosaics (Lucieer et al., 2014; Turner et al., modeling of ionospheric and tropospheric errors (Snay and Soler, 2014a,b; Zabala, 2017; Martínez-Carricondo et al., 2018). The photo- 2008). PRS network data can be used in real time or in post-processing grammetric software makes it possible to use automatically geotagged treatment. image PC to locate the individual images (DG) or set GCP to aero- triangulate the block (IG). In the case of geotagged image PC (post- 2.2. Materials processed or not) and IMU data (if provided), the photogrammetric software locates the frames in the map space, following the provided The UAV platforms used were an octocopter (Fig. 4a) and a hex- external position coordinates (X,Y,Z), and finds the optimal orientation acopter (Fig. 4b). Both had a Pixhawk controller, which was assembled angles (ω,φ,κ) between overlapped images by rotating, translating and by the HEMAV Company to provide the team with more functionality scaling the individual frames. We followed the photogrammetric soft- and additional security measures (Table 2). ware workflow proposal developed by the United States Geological The octocopter was used to carry the SONY Alpha 7RII Digital Survey (USGS) National Unmanned Aircraft Systems (UAS) Project Single-Lens Reflex (DSLR) camera (Table 3), which is sensitive in Red, Office (NUPO) for DSLR imagery (USGS-NUPO(a) (United States Green and Blue (RGB) spectral regions, and included a 15-mm fixed Geological Survey - National Unmanned Aircraft Systems Project lens (Voightländer, 2017). A memory card was located in the flash Office), 2017) and MicaSense RedEdge multispectral imagery (USGS- hotshoe, which recorded the GNSS time at the flash firing. The syn- NUPO(b) (United States Geological Survey - National Unmanned chronization between the shooter rear curtain and flash firing is 1/ Aircraft Systems Project Office), 2017) for the purpose of obtaining 3D 250th of a second (Sony, 2017); the distance travelled by the UAV reconstruction from UAV photogrammetric imagery using Structure within this time at the programmed velocity (5 m/s) is 0.02 m. from Motion (SfM) techniques. A 3D point cloud was created, from The hexacopter was used to carry the multispectral and multilens which a Digital Surface Model (DSM) was generated. When the optimal MicaSense RedEdge camera (Table 4), which is sensitive to the RGB, correlation of homologous tie points was found to account for the red-edge and near-IR spectral regions (MicaSense, 2015). A port spe- perspective, an automatic camera calibration was achieved with the cifically intended for this purpose facilitated connection and synchro- current environmental conditions, which used more points (> 1k) than nization between the camera and the GNSS device for recording the in classic lab calibrations (Harwin et al., 2015). Then, the final geor- time in a memory card and geotagging the images. eferenced (UTM31 N ETRS89) orthophotomosaic at 0.025 m SR in Sony The sensors were mounted on the gimbal of the corresponding RGB imagery and at 0.05 m SR in MicaSense RedEdge multispectral platform. In the lab, the camera nodal point was aligned with the imagery (Table 5) was exported to a remote sensing and GIS software gimbal axis to prevent distortions in the lever-arm offsets when the (Microstation v 8i (Bentley Systems, 2017) and MiraMon v 8.01b (Pons, gimbal compensates for the platform attitude angles. The lever-arm 2016)) so as to compare the image-derived coordinates to surveyed offset (X,Y,Z) between the Perspective Center (PC) camera position and points. We used a different camera for one of the experiments in order the Antenna Phase Center (APC) was measured at a sub-centimetric to carry out a more thorough exploration in spectral and spatial re- level. The constant displacement vector (Skaloud et al., 1996) was solutions. We are aware that using two cameras could introduce further introduced in the post-processing software to compensate for asymmetry in the experiment design. Nevertheless, we consider the the lever-arm offsets. Based on the corrected onboard GNSS signal and varying results to substantially enrich the comparison part of the paper. the time of image sensing, the PPK software interpolates the position of the camera PC camera position at the time of exposure, outputting the 2.3. Methods accurate estimation of every image PC position. The photogrammetric software (Agisoft LLC, 2018) used in our The position of the GCPs was obtained by applying two Fig. 2. Sky view factor map obtained from the Digital Surface Model (lidar official cartography (SR = 0.25 m) (ICGC, 2018) and the position of the Ground Control Points. 133 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 Fig. 3. a) The red rectangle indicates the distribution of surveyed GCPs on the orthophotomosaic generated with the PPK process (SR = 0.025 m). Background (outside the red rectangle): Official state-agency cartography (SR = 0.25 m) (ICGC, 2018). b) The DGNSS receiver, which was held on a tripod with a clamp, received measurements that were later analyzed in post-processing (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article). Fig. 4. a) Octocopter platform with the RGB sensor mounted on the gimbal; b) Hexacopter platform with the multispectral sensor mounted on the gimbal. 134 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 Table 2 UAV platforms and their main features. Manufacturer-Model Diameter Size (cm) Flight autonomy (min) MTOW (kg) Optical sensor embedded type DJI-S1000 105 15 11 RGB DJI-S900 90 18 8.2 Multispectral Table 3 SONY Alpha 7RII camera and its main features. Manufacturer-Model Size (mm) Weight (g) Raw radiometric resolution (bits) and expanded dynamic range (DN) SONY Alpha 7RII 127 × 96 × 60 625 12 (256 × 3 bands) Sensor type Sensor size (pixels) Pixel size (μm) Focal Length (mm) Bands 35 mm-CMOS 7952 × 5304 4.53 15 R-G-B methodologies: firstly, all GCP measurements were made using static high variability of NSSDA estimations is a result of the GCPs' sample RTK, and secondly, the static DGNSS was set to collect data for a size and population deviation (Ariza and Atkinson, 2008). In our case minimum of 20 min, which were post-processed (PP) to determine the study, the relatively small sample size due to the limited extension of most exact GCP position possible. Moreover, we opted to use RTK in the study area led to a more conservative estimation, although the low order to expediently obtain coordinates that would be very similar to population deviation seems to indicate otherwise. As our sample size is those calculated in the PP, thereby facilitating confirmation that the smaller than that of the Ariza and Atkinson (2008) simulations, we results obtained from the two methods are consistent because a priori worked with NSSDA estimations. Indeed, for larger study areas and/or PP measurements are much more reliable. In this study, we can see that with more GCPs, those criteria proposed by Ariza and Atkinson could be the difference between the RTK and PP data is less than we expected, as applied. it is horizontally almost always below 0.01 m, and vertically below The design basically consisted of checking the accuracy of the or- 0.04 m (Table 6). However, for Point-7 there are larger variations be- thophotomosaics generated with the photogrammetric software using tween the two methods resulting in differences in X and Y of 0.06 m and four different georeferencing methods, taking well-defined test points 0.09 m, and 0.14 m in elevation. on the terrain as references, and comparing the results to the ASPRS Hence, the Point-7 position was considered not reliable enough to standards. Finally, we proceeded to evaluate the differences between be taken as a reference. Thus, from the initial eight GCP we planned to them, whereby we considered the estimated economic costs that can be use, we ultimately worked with the seven points that had well-known amortized in several flight campaigns. coordinates. Mean differences are 0.002 m in easting, 0.006 m in northing and −0.024 m in elevation (without considering Point-7). (1) GNSSNAV: The onboard navigation GNSS receiver used to de- The UAV georeferencing methods evaluation was carried out fol- termine the platform position works with the GPS/GLONASS con- lowing the new ASPRS Positional Accuracy Standards for Digital stellations (using the European Geostationary Navigation Overlay Geospatial Data, which addresses recent innovations in digital imaging System (EGNOS) augmentation system), L1/L2 frequencies with 20 and non-imaging sensors, airborne GNSS, IMU and AT technologies channels, and a 10 Hz frequency rate (10 timestamps per second). (ASPRS, 2015). This standard defines accuracy classes based on RMSE The navigation GNSS method automatically geotags the central thresholds for digital orthoimagery, digital planimetric data, and digital position (X,Y,Z) of the frames by recording the coordinates of the elevation data (ASPRS, 2015). Horizontal accuracy was assessed using onboard GNSS APC at the closest trigger time (the offset between RMSEX for the horizontal linear RMSE in the X direction (Easting), the APC and the PC was not accounted for), accordingly to the RMSEY for the horizontal linear RMSE in the Y direction (Northing), 10 Hz rate without differential correction. As previously in- and RMSEr for the horizontal linear RMSE in the radial direction, in- troduced, at a velocity of 5 m/s, the positioning error due to this cluding both X- and Y-coordinate errors. Vertical accuracy was assessed gap can range 0.5 m. In the photogrammetric software processing, using RMSEZ for the vertical linear RMSE in the Z direction (elevation). we applied the workflow recommended by USGS for DSLR imagery Note that vertical accuracy was evaluated as Non-Vegetated Area (USGS-NUPO(a) (United States Geological Survey - National (NVA) RMSE statistics, since the GCPs were placed in vegetation-free Unmanned Aircraft Systems Project Office), 2017). The derived locations. The threshold used to account for 95% of the error in the one- orthophotomosaic absolute accuracy was validated with all the dimensional Z-axis over NVA was ≤1.96 x RMSE (ASPRS, 2015). surveyed ICPs. The approximate economic cost of the GNSS device Horizontal and vertical thresholds (assuming normal distribution) is less than 50 €. If direct positioning of images is applied, no time established by ASPRS following the methodologies of the National investment is necessary to process the onboard GNSS data. How- Standard for Spatial Data Accuracy (NSSDA) FGDC, 1998) are shown in ever, the photogrammetric software processing itself can take hours Table 7 (Table 7). to determine the bundle block adjustment. Two PPK variants were According to previous studies, in a simulation of populations, the used, which involved different hardware and fieldwork costs: Table 4 MicaSense RedEdge sensor and its main features. Manufacturer-Model Size (mm) Weight (g) Raw radiometric resolution (bits) and expanded dynamic range (DN) MicaSense RedEdge 121 × 66 × 46 150 12 (0–65,535) Sensor type Sensor size (pixels) Pixel size (μm) Focal Length (mm) Number of bands CCD 1280 × 960 3.75 5.5 5 #1 Blue FWHM (nm) #2 Green FWHM (nm) #3 Red FWHM (nm) #4 Red-edge FWHM (nm) #5 NIR FWHM (nm) 468–491 548–568 666–676 712–723 814–865 135 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 Table 5 Flight plan main features. Platform and sensor Side overlap (%) Forward overlap (%) Flight height (m) Spatial resolution (m) Georeferencing method Octocopter 60 80 80 0.025 GNSSNAV; PPK2; GCP-ICP -RGB Hexacopter 60 80 80 0.050 PPK1 Multispectral Table 6 Target coordinates (reference Ground Control Points) measured in the static Post-Processed (PP), and residuals with respect to Real-Time Kinematic (RTK) mea- surements. TARGET COORDINATES RESIDUALS (m) UTM-31 N ETRS89 (m) POST-PROCESSED (PP) PP vs. RTK Point-ID X (Easting) Y (Northing) Z (Elevation) Δ X (Easting) Δ Y (Northing) Δ Z (Elevation) Point-1 378504.631 4645236.608 725.277 0.005 −0.001 −0.026 Point-2 378469.154 4645250.728 725.712 0.008 0.005 −0.012 Point-3 378366.020 4645275.324 726.314 0.008 0.006 −0.019 Point-4 378394.089 4645206.539 729.967 0.005 0.006 −0.014 Point-5 378385.341 4645145.964 734.201 −0.006 0.012 −0.037 Point-6 378435.762 4645185.649 728.625 −0.006 0.010 −0.036 Point-7 378488.194 4645271.260 723.034 −0.060 0.087 0.141 Point-8 378578.130 4645286.430 719.098 −0.004 0.005 −0.021 Table 7 coordinates. Although PPK2 can work with a PRS, as PPK1 does, we Horizontal and Vertical (NVA) Accuracy Standards for Geospatial Data used to nevertheless used a station located at known coordinates in the evaluate the four positioning methods. study area to improve the local characterization of ionospheric, Horizontal Accuracy at Horizontal Accuracy at Vertical Accuracy at 95 % tropospheric and geometric errors affecting the L1/L2 carrier-phase 95 % Confidence X, Y 95 % Confidence radial Confidence Z (NVA) GNSS signals. Thus, we were able to minimize the distance-de- pendent errors affecting the accuracy. The function of this base, ≤ 2.4477 x RMSE ≤ 1.7308 x RMSEr ≤ 1.9600 x RMSE which was linked to the nearest PRS, was to receive differential corrections continuously to further correct the onboard GNSS po- sition. In the photogrammetric software processing, we applied the (2) PPK1: PPK1 method (Rokubun, 2018) consists of retrieving the PC workflow recommended by USGS for DSLR imagery (USGS-NUPO camera position at the exposure time by calculating the flight path (a) (United States Geological Survey - National Unmanned Aircraft and interpolating the acquisition time of each image on the calcu- Systems Project Office), 2017). The final orthophotomosaic abso- lated trajectory. Argonaut 10 Hz single-frequency carrier-phase lute accuracy was validated with seven surveyed GCPs. Although GNSS receiver with a 9-degree IMU embedded was used to obtain this method does not rely on any GCP, it is nevertheless dependent the position of the UAV. The triggering position was given by the on the base station located in the study area. The approximate cost onboard GNSS time sent to the camera MicroSD port, which was of the rover module itself was about 4000 € and the base module designed for this purpose. A Software as a Service (SaaS) (Rokubun, was about 4500 € (Drobit, 2017). 2018) was used for data post-processing and determining the (4) GCP–ICP: This method is not supported by onboard GNSS data. IG, camera pose at exposure time, whereby we transferred the images which is a well-known technique based on AT, provides a bundle and the MicroSD card data to the company by following online block adjusted to the ground references, and can therefore de- steps prescribed on their website (Rokubun, 2018). The PPK1 termine the EOP of each image block in an indirect way. The image method, which requires the support of a PRS that continuously orientation with photogrammetric software was done without using receives differential corrections and further corrects the onboard the imagery geotagged information. Rather, we manually located GNSS position, does not rely on any GCP or ground station in the the GCP markers in their corresponding individual image. We study area. In the photogrammetric software processing, we applied performed cross-validation to validate the final orthophotomosaic’s the workflow recommended by USGS for multispectral imagery absolute accuracy, thereby generating seven orthophotomaps (USGS-NUPO(b) (United States Geological Survey - National georeferenced with six GCPs; the remaining point was the ICP Unmanned Aircraft Systems Project Office), 2017). The final or- evaluated in each orthophotomap. Cross-validation (leave-one-out) thophotomosaic absolute accuracy was validated with all surveyed consists of estimating a model from a subset, leaving an isolated GCPs. The approximate cost of the Argonaut receiver was 350 €, value (test point) to perform the validation process, which is then while the SaaS was 200 € (Rokubun, 2018). carried out iteratively with all the data of the set (Picard and Cook, (3) PPK2: PPK2 method (Drobit, 2017) proved to be the most sophis- 1984). Assuming an accessible region of interest, the economic ticated solution we tested and evaluated. The synchronization be- costs of materializing GCPs, including renting a static DGNSS and tween the camera triggering and the GNSS time (Rehak and the personnel costs approximately amounted to 500 € The time Skaloud, 2017) was carried out by a card device located in the flash invested was about one hour per GCP for painting and materializing support slit. Said card device collects the flash firing signal just concrete platform points; obviously, the time duration would be before the rear curtain closes (flash is activated and synchronized reduced if tarps were used or stable ground locations were painted. with the rear curtain shutter). Drobit v1.6 software (Drobit, 2017), In our study, we took one hour for each GCP static GNSS mea- which was based on RTKlib open source libraries, was used for data surement, two hours for post-processing static measurements and processing in PPK2. During the post-processing stage, the lever-arm two processing hours for accurately locating GCPs in individual offsets were introduced in the software to obtain PC, instead of APC images. 136 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 Fig. 5. Methodology for obtaining the image horizontal coordinates of GCPs (example for Point-3). The coordinates in the resulting orthophotomosaics (image co- the method in direct georeferencing of UAV imagery of very high re- ordinates) were checked by two operators separately (Operator A and solution. In fact, in other studies using NSSDA accuracy standards Operator B). In order to accurately determine the center of the target, FGDC, 1998), we found clearly larger values when scale and pixel di- i.e. the rod in the center of the concrete base, the procedure was based mension are compared. on digitizing the squared border (50 cm side) of the checkered area and The one-dimensional indicator of accuracy with respect to the re- drawing a centered cross. The coordinate of the crossing lines node was ference points was positively checked for normality. As expected, the taken as the surveying rod coordinate in the image. Operator A per- geometric errors in the univariate space (RMSE) were also of different formed this procedure twice, while operator B did a third round on the orders of error magnitude between the different methods (Table 8). In extraction of image coordinates. The mean value of the three manual the GNSSNAV option, the RMSE was under 1 m in X, 0.5 m in Y and rounds was the image value (see Table S1 Supplementary materials) over 4 m in Z, i.e. in the metric order; in the PPK1 option, the RMSE was compared with the reference surveyed coordinates. The image co- up to 0.15 m in X, 0.20 m in Y and 0.24 m in Z, i.e. in the decimeter ordinates were considered valid if 90% of the coordinates have a dif- order; in the PPK2 option, the RMSE was up to 0.025 m in X or Y; and ference from the mean value less than one-third of the pixel size and the 0.036 m in Z; and in the GCP–ICP option the RMSE was up to 0.014 m in remaining 10% do not exceed half of the pixel size; otherwise, mea- X, 0.018 m in Y and 0.030 m in Z, i.e. in the centimetric order (Table 9). surements had to be repeated. In the GCP-ICP cross-validation proce- The synthetic indicator of planimetry precision, which assumed a dure, seven orthophotomaps were made, whereby one GCP was left out normal distribution of the geometric errors in the XY bivariate space in each map, and Operators A and B extracted the coordinates of this (combined in X and Y, or radial), showed a statistical error in 95% of point with the same described procedure (Fig. 5). the cases, i.e. 1.837 m in the GNSSNAV method, 0.443 m in the PPK1 method, 0.062 m in the PPK2 method, and 0.039 m in GCP-ICP method (Table 9). The radial indicator proved to be a robust method for 3. Results determining the planimetric error, and thus for estimating the SR useful for each method, although the bivariate standard deviation of the errors The well-known surveyed coordinates of the GCP points (static was very valuable in obtaining the planimetric accuracy order of error DGNSS with post-processing) were compared with their estimated magnitude of each method (Table 9). image-derived position for each method (Table 8). Calculation of one- dimensional (Easting, Northing or elevation) differences between the reference and the estimated values shows that the GNSSNAV method 4. Discussion has the highest inaccuracies at all points, whereby PPK1 reduces the order of the error magnitude at all points, while GCP–ICP and PPK2 In this applied study, we used a lower number of GCPs than the have very similar levels of accuracy at Point-3, but has clearly higher number used in previous geometric accuracy evaluation studies (Ariza differences in the remaining points (Table 8). It is worth noting that the and Atkinson, 2008; Martínez-Carricondo et al., 2018). All the methods PPK2 results show a mean bias (similar to the pixel dimension of we tested have been proven to perform well in controlled situations. 0.025 m), but these values are not large enough to invalidate applying However, for this particular study, we used a realistic setting in UAV Table 8 Residuals between reference coordinates (PP) and coordinates obtained using UAV positioning methods: (1) GNSSNAV, (2) PPK1, (3) PPK2, (4) GCP–ICP. Positioning method Point-ID Point-1 Point-2 Point-3 Point-4 Point-5 Point-6 Point-8 μ σ 1) Residuals (m) PP vs. GNSSNAV Δ X −0.946 −0.906 −0.845 −0.936 −1.001 −0.980 −0.938 −0.936 0.050 Δ Y 0.436 0.476 0.604 0.528 0.511 0.511 0.398 0.495 0.067 Δ Z −4.257 −4.041 −3.591 −4.226 −4.652 −4.515 −4.095 −4.197 0.345 2) Residuals (m) PP vs. PPK1 Δ X 0.082 0.124 0.262 0.187 0.114 0.112 0.081 0.137 0.065 Δ Y −0.265 −0.272 −0.180 −0.116 −0.083 −0.219 −0.237 −0.196 0.074 Δ Z 0.305 0.262 0.026 0.188 0.301 0.279 0.178 0.220 0.100 3) Residuals (m) PP vs. PPK2 Δ X −0.030 −0.015 0.004 −0.016 −0.024 −0.034 −0.038 −0.022 0.014 Δ Y 0.018 0.026 0.025 0.014 0.032 0.027 0.030 0.025 0.060 Δ Z −0.044 −0.010 −0.029 −0.032 0.028 −0.046 −0.045 −0.025 0.027 4) Residuals (m) PP vs. GCP-ICP cross-validation Δ X −0.012 0.008 0.013 0.007 0.007 −0.020 −0.023 −0.003 0.015 Δ Y −0.020 −0.002 −0.021 −0.008 −0.025 0.009 −0.025 −0.013 0.013 Δ Z 0.017 0.023 −0.029 −0.030 0.022 −0.035 −0.046 −0.011 0.030 137 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 Table 9 agency at 2.5 m (ICGC, 2018; PNOA, 2018) or satellite sensors with RMSE in each of the univariate spatial components (X, Y and Z) and in the coarser SR, such as the 10-m bands of MSI onboard Sentinel-2 (ESA, bivariate planimetric component (radial, combined XY) obtained using (1) 2015), or the 15-m panchromatic bands of ETM + onboard Landsat-7 GNSSNAV, (2) PPK1, (3) PPK2 and (4) GCP–ICP. and OLI onboard Landsat-8 (NASA, 2018). RMSE RMSE (95 % cases) The PPK1 method, which improved the raw GNSS data by one order of error magnitude without in situ ground support, approximately costs 1) RMSE (m) GNSSNAV X 0.937 1.837 600 € (amortizable). Using PRS increases the accuracy due to the cor- Y 0.499 0.977 R (radial) 1.062 1.837 rection of the GNSS satellite signal. However, a post-processing service Z 4.209 8.249 has to be additionally contracted. This option was accurate enough to 2) RMSE (m) PPK1 X 0.150 0.294 combine UAV resampled imagery with the airborne-acquired products Y 0.207 0.406 of official mapping agency at 0.25 m or 0.50 m SR, lidar data at 2 R (radial) 0.256 0.443 points/m2(ICGC, 2018; PNOA, 2018), or sub-metric satellite imagery as Z 0.238 0.467 3) RMSE (m) PPK2 X 0.025 0.050 the panchromatic band of WorldView-4 (Nadir: 0.31 m SR, 56° Off- Y 0.025 0.050 Nadir: 1.00 m SR) (DigitalGlobe, 2018). However, it did not reach the R (radial) 0.036 0.062 goal of registering UAV imagery with SR > 0.05 m. Moreover, this Z 0.036 0.070 method is not useful if PRS is not accessible within a radius of 20 km. 4) RMSE (m) GCP–ICP X 0.014 0.028 Y 0.018 0.035 The PPK2 method, which approximately costs 8500 € (amortizable) R (radial) 0.023 0.039 was consistent enough in generating large-scale mapping from UAV Z 0.030 0.059 imagery with less effort than the GCP–ICP method. The use of a station in the study area and a double-frequency receiver allows a better cor- rection of satellite signal, and the synchronization of camera acquisition operational cases. Moreover, in our study, we learned that the field with the GNSS receiver is almost optimal. The cartographic quality that static measurements might result in proving the planned location of one can be generated will depend on the specific norms applied. GCP to be not reliable enough, such that this would have to be dis- Nevertheless, it is clear that highly detailed maps (ASPRS, 2015; FGDC, carded, as in the case of Point-7. 1998) can be produced with the PPK2 method. According to ASPRS Among the methods studied, the GCP-ICP method yielded the tables (see Table 2), the goal of mapping at 1:200 scale with 5-cm SR highest accuracy altogether. The PPK2 method, the PPK1 method and imagery can be achieved in planimetry and in altimetry. Moreover, it the GNSSNAV method yielded the second, the third and the fourth seems that the UAV products processed with the PPK2 method con- highest accuracies, respectively. Regarding the residuals (Table 8) and stitute a valid alternative to the GCP–ICP method. However, such ac- the image pixel size as a reference value, a clear bias is present in curacy depends on the logistics of mounting a base station and in situ GNSSNAV and PPK1. These methods are not only less accurate in ground support. general terms, but they also provide a systematic displacement due to Among all four methods studied, the GCP-ICP method proved to be the combined error sources of the lever-arm offset, the lack of any the most accurate method. As far as the time invested and the economic ground reference point (GNSSNAV), or that this point was not close costs involved, we spent one hour for static GNSS measurements for enough to our reference value (PPK1). Interestingly, the PPK2 and each GCP (displacing and setting the DGNSS) with specialized equip- GCP–ICP methods, which had one or more ground reference points in ment (renting a DGNSS) and hiring a skilled personnel (500 €), and the same study area, provided acceptable biases when compared with about two hours for post-processing the data. If GCPs materialized in the reference value. Nevertheless, the economic cost, invested time and concrete bases prove useful in future campaigns, as was the case in our ground requirements of each method are different and can fit several study, then the process of materializing and painting the targets should remote sensing purposes. Note that even UAV imagery can be acquired be carried out the day before the flight. In our experience, the markers at > 10 cm SR, and can thus be resampled to coarser pixel size to ab- had to be placed in the corresponding images, which proved to be a sorb geometric uncertainty, which is typically the radial error. Re- tedious task that required over one hour. Apart from setting the GCPs, it sampled imagery, which can be used to improve the processing of other is fair to note that the region of interest is not always accessible and this remote sensing products, is not as detailed as drone-acquired imagery method is not useful in every case. (Table 10). The GNSSNAV method provided the poorer planimetric and alti- metric accuracies, by overpassing 1 m. However, at an approximate cost 5. Conclusions of 50 €, GNSSNAV was the most affordable and the easiest-to-use op- tion, and did not require ground or external support. Although this The positioning of UAV-acquired frame imagery can be solved by option is not useful for combining UAV imagery with satellite or air- applying different methods with corresponding accuracies (ASPRS), borne imagery < 2 m SR, it can nevertheless be useful for analyzing the economic costs and time investments. Aside from the multiple combi- intra-pixel features of airborne-acquired products of official mapping nations of optical sensors, GNSS receivers, platforms, gimbals, and Table 10 Summary of the pros and cons of the evaluated georeferencing methods. 1) GNSSNAV 2) PPK1 3) PPK2 4) GCP–ICP RMSEr (m) 1.062 0.256 0.036 0.023 RMSEz (m) 4.209 0.238 0.036 0.030 Approximate initial investment cost (€) 50 600 8500 500 Time consuming (h) ≥1 ≥24 ≥24 ≥48 Involved personnel ≥1 ≥1 ≥2 ≥2 In situ ground support NO NO YES YES Suggested SR of the corrected imagery (m) ≥2 ≥0.25 ≥0.05 ≥0.025 Examples of RS products to be combined Sentinel-2, imagery and WorldView imagery, airborne AHS, CASI, UAV Sequoia, Cubert data UAV reflex cameras data with UAV data coarser lidar data, and coarser and coarser and coarser ASPRS associated maximum map scale 1:12000 1:1500 1:400 1:200 138 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 software that can be tested, we compared four commonly used geor- Barcelona. eferencing solutions in a real operational case for environmental remote sensing purposes with commonly used materials. Although the raw Appendix A. Supplementary data GNSS receiver direct georeferencing solution is not suitable for re- gistering UAV data in different flight acquisitions, it is easy to use, can Supplementary material related to this article can be found, in the be purchased at a low cost, and does not require ground support, which online version, at doi:https://doi.org/10.1016/j.jag.2018.10.018. is useful for visual inspections of remote sensing products with a pixel size coarser than 2 m, such as Landsat-like imagery. The Post-Processed References Kinematic solution using single-frequency carrier-phase GNSS receiver and correcting GNSS signal with information from a permanent re- Aasen, H., Honkavaara, E., Lucieer, A., Zarco-Tejada, P.J., 2018. Quantitative remote ference station located up to 20 km away, is valid for registering drone sensing at ultra-high resolution with UAV spectroscopy: a review of sensor tech- nology, measurement procedures, and data correction workflows. Remote Sens. 10, imagery with airborne lidar data or orthoimagery with a ground sam- 1091–1933. https://doi.org/10.3390/rs10071091. pling distance of up to 25 cm. The Post-Processed Kinematic solution Agisoft LLC, 2018. PhotoScan User Manual, Professional Edition, Version 1.4.1. URL: using double-frequency carrier-phase GNSS receiver and correcting the (Last date accessed: 1 Oct. 2018). http://www.agisoft.com/pdf/photoscan-pro_1_4_ en.pdf. GNSS signal with information from a base station located in the study Ariza, F.J., Atkinson, A.D., 2008. Variability of NSSDA estimations. J. Surv. Eng. 134, area, reaches the goal of registering UAV imagery at its common spatial 39–44. https://doi.org/10.1061/(ASCE)0733-9453(2008)134:2(39). resolution (≥5 cm). However, the logistics require the support of in situ ASPRS (American Society for Photogrammetry and Remote Sensing), 2014. ASPRS Positional Accuracy Standards for Digital Geospatial Data. Edition 1, Version 1.0.0. materials and the cost is higher. The conventional indirect georefer- URL: (Last date accessed: 1 Oct. 2018).. https://www.asprs.org/wp-content/ encing using ground control points is the most accurate solution, as it is uploads/2015/01/ASPRS_Positional_Accuracy_Standards_Edition1_Version100_ useful for registering UAV imagery with any other remote sensing November2014.pdf. ASPRS (American Society for Photogrammetry and Remote Sensing), 2015. New coarser product. However, it also requires intense in situ ground sup- Standards for New Era: Overview of the 2015 ASPRS Positional Accuracy Standards port and is by far more time-consuming. for Digital Geospatial Data. URL: (Last date accessed: 1 Oct. 2018. https://www. This study demonstrates that different UAV optical frame imagery asprs.org/wp-content/uploads/2015/01/PERS_March2015_Highlight.pdf. georeferencing methods can be suitable and most effective when com- Bentley Systems, 2017. MicroStation, Version 8.1. URL: (Last date accessed: 1 Oct. 2018). https://www.bentley.com. bined with conventional and freely available remote sensing products Bisnath, S., Wells, D., Santos, M., Cove, K., 2004. Initial results from a Long baseline, (e.g. Landsat, Sentinel-2, official mapping agencies lidar or orthoima- kinematic, differential GPS carrier phase experiment in a Marine environment. gery). Obviously, the most accurate solutions are more expensive. PLANS 2004 Position Location and Navigation Symposium. IEEE, Monterey, USA, pp. 625–631. https://doi.org/10.1109/PLANS.2004.1309052. 2004 Apr 26–29, Cat. No. Moreover, time and ground support are important factors to consider. 04CH37556. Our paper is intended to be a quick and reliable guide for environ- Bláha, M., Eisenbeiss, H., Grimm, D., Limpach, P., 2012. Direct georeferencing of UAVs. mental scientists that provides a comparison of four known georefer- ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXVIII-1/C22, 2011 ISPRS Zurich 2011 Workshop; encing methods to help them select the best option that suits their 2011 Sep 14–16; Zurich, Switzerland. p. 131–136. https://doi.org/10.5194/ispr- purpose, particularly in view of the accuracy required when combining sarchives-XXXVIII-1-C22-131-2011. UAV imagery with well-known airborne and satellite remote sensing BOE (Boletín Oficial del Estado), 2017. “Real Decreto 1036/2017, de 15 de diciembre, por el que se regula la utilización civil de las aeronaves pilotadas por control remoto” platforms. [Royal Decree RD 1036/2017, of December 15, for the regulation of the civil use of remotely piloted aircrafts], vol. 316. Ministerio de Fomento, Madrid, Spain, pp. Author contributions 129609–129641. BOE n URL: (Last date accessed: 1 Oct. 2018). http://www.boe.es/ boe/dias/2017/12/29/pdfs/BOE-A-2017-15721.pdf. Borgogno Mondino, E., Gajetti, M., 2017. Preliminary considerations about costs and F.M., J.P. and C.P. conceived and designed the experiments. J.P and potential market of remote sensing from UAV in the Italian viticulture context. Eur. J. C.P performed the field data experiments. J.P and C.P. analyzed the Remote Sens. 50, 310–319. https://doi.org/10.1080/22797254.2017.1328269. data. J.P., C.P. and X.P. wrote the paper. F.M and X.P. supervised the Chio, S.-H., 2016. VBS RTK GPS-assisted self-calibration bundle adjustment for aerial triangulation of fixed-wing UAS images for updating topographic maps. Bol Cienc paper and the overall experiments. Geod. 22, 665–684. https://doi.org/10.1590/S1982-21702016000400038. Chio, S.-H., Lin, C.-H., 2017. Preliminary study of UAS equipped with thermal camera for Conflicts of interest volcanic geothermal monitoring in Taiwan. Sensors 17, 1649–1666. https://doi.org/ 10.3390/s17071649. Colomina, I., Molina, P., 2014. Unmanned aerial systems for photogrammetry and remote The authors declare that there are no conflicts of interest. The sensing: a review. ISPRS – J. Photogramm. 92, 79–97. https://doi.org/10.1016/j. sponsors had no role in the design of the study, the collection, analyses, isprsjprs.2014.02.013. Crommelinck, S., Bennett, R., Gerke, M., Ying Yang, M., Vosselman, G., 2017. Contour or interpretation of data, the writing of the manuscript, or the decision detection for UAV-Based cadastral mapping. Remote Sens. 9, 171–184. https://doi. to publish the results. org/10.3390/rs9020171. Daakir, M., Pierrot-Deseilligny, M., Bosser, P., Pichard, F., Thom, C., Rabot, Y., Martin, O., 2017. Lightweight UAV with on-board photogrammetry and single-frequency GPS Acknowledgments positioning for metrology applications. ISPRS – J. Photogramm. 127, 115–126. https://doi.org/10.1016/j.isprsjprs.2016.12.007. This work was supported by the European Union through the DigitalGlobe, 2018. DigitalGlobe Products and Services. URL: (Last date accessed: 25 Jan. 2018. https://www.digitalglobe.com/#products. ECOPOTENTIAL Project (H2020 641762-2 EC) and by an Innovation Drobit, 2017. The Onboard Surveyor. A PPK Solution for Drone-mapping. URL: (Last date grant from the Spanish Ministry of Science (ACAPI, CGL2015-69888-P). accessed: 1 Oct. 2018). https://www.drobit.es/drobit-specs-sheet.pdf. The GRUMETS Research Group is partially supported by the Catalan ESA (European Space Agency), 2015. Sentinel-2A User Handbook. Released 24/07/2015. Government under Grant (SGR2017-1690). C.P. is a recipient of a FI- Rev.2 https://sentinels.copernicus.eu/web/sentinel/user-guides/document- library/-/asset_publisher/xlslt4309D5h/content/sentinel-2-user-handbook URL: DGR scholarship grant (2016B_00410). X.P. is a recipient of an ICREA (Last date accessed: 1 Oct. 2018). Academia Excellence in Research Grant (< GN542016-2020 < / FGDC (Federal Geographic Data Committee), 1998. Geospatial Positioning Accuracy GN4 >). We would like to acknowledge R.F., who collaborated on Standards. Part 3: National Standard for Spatial Data Accuracy (NSSDA). URL: (Last date accessed: 1 Oct. 2018). https://www.fgdc.gov/standards/projects/accuracy/ measuring the control points with the static GNSS and their post-pro- part3/chapter3. cessing, X.C., who collaborated on the materialization of points and Harwin, S., Lucieer, A., Osborn, J., 2015. The impact of the calibration method on the logistics, and J.B., who also collaborated on the materialization of the accuracy of point clouds derived using unmanned aerial vehicle multi-view ste- reopsis. Remote Sens. (Basel) 7, 11933–11953. https://doi.org/10.3390/ points. The HEMAV Company collaborated on the flight campaign and rs70911933. made its technical team available. M.S.'s expertise in the UAV flight was ICGC (Institut Cartogràfic i Geològic de Catalunya), 2017. Fonaments Del Posicionament of great help. This work was carried out as part of the Doctoral Program Diferencial I Introducció a NTRIP” [Fundamentals of Differential Positioning and Introduction to NTRIP]. URL: (Last date accessed: 1 Oct. 2018). http://www.icgc. of the Geography Department of the Universitat Autònoma de 139 J.-C. Padró et al. Int J Appl Earth Obs Geoinformation 75 (2019) 130–140 cat/content/download/48729/337223/version/5/file/NTRIP_a_ICGC_%20v8.pdf. Information Sciences, Volume II-1/W1, 2015 International Conference on Unmanned ICGC (Institut Cartogràfic i Geològic de Catalunya), 2018. Aerial Photos and Orthophotos. Aerial Vehicles in Geomatics; 2015 Aug 30–Sep 02 23–30. https://doi.org/10.5194/ URL: (Last date accessed: 1 Oct. 2018. http://www.icgc.cat/en/Public- isprsannals-II-1-W1-23-2015. Administration-and-Enterprises/Downloads/Aerial-photos-and-orthophotos. Remondino, F., Barazzetti, L., Nex, F., Scaioni, M., Sarazzi, D., 2012. UAV Ihde, J., Habrich, H., Sacher, M., Söhne, W., Altamimi, Z., Brockmann, E., Bruyninx, C., Photogrammetry for Mapping and 3d Modeling -current Status and Future Caporali, A., Dousa, J., Fernandes, R., Hornik, H., Kenyeres, A., Lidberg, M., Mäkinen, Perspectives-. ISPRS - International Archives of the Photogrammetry, Remote Sensing J., Poutanen, M., Stangl, G., Torres, J.A., Völksen, C., 2014. EUREF’s contribution to and Spatial Information Sciences, Volume XXXVIII-1/C22, 2011 ISPRS Zurich 2011 national, European and global geodetic infrastructures. Proceedings of the IAG Workshop; 2011 Sep 14–16, Zurich, Switzerland, pp. 1–7. https://doi.org/10.5194/ General Assembly 139, 189–196. https://doi.org/10.1007/978-3-642-37222-3_24. isprsarchives-XXXVIII-1-C22-25-2011. 2011 Jun 28–Jul 2. Rokubun, 2018. Argonaut Datasheet. URL: (Last date accessed: 1 Oct. 2018). http:// Komárek, J., Klouček, T., Prošek, J., 2018. The potential of Unmanned Aerial Systems: A rokubun.cat/wp-content/uploads/2017/12/ARGONAUT_Brochure.pdf. tool towards precision classification of hard-to-distinguish vegetation types? Int J Skaloud, J., Cramer, M., Schwarz, K.P., 1996. Exterior orientation by direct measurement Appl Earth Obs. 71, 9–19. https://doi.org/10.1016/j.jag.2018.05.003. of camera position and attitude. ISPRS - International Archives of Photogrammetry Geosystems, Leica, 2017. Leica Viva GS14 Technical Specifications. URL: (Last date ac- and Remote Sensing Vol. XXXI, Part B3; 1996 Jul 9–19; Vienna, Austria. P. 125–130. cessed: 1 Oct. 2018). http://www.leica-geosystems.es/downloads123/zz/gpsgis/ URL: (Last date accessed: 1 Oct. 2018). http://www.ifp.uni-stuttgart.de/ VivaGS14/brochures-datasheet/Leica_Viva_GS14_DS_en.pdf. publications/1996/wien96_gpsins.pdf. Lin, A.Y.M., Novo, A., Har-Noy, S., Ricklin, N.D., Stamatiou, K., 2011. Combining Snay, R.A., Soler, T., 2008. Continuously operating reference stations (CORS): history, GeoEye-1 satellite remote sensing, UAV aerial imaging, and geophysical surveys in applications, and future enhancements. J Surv Eng. 134, 95–104. https://doi.org/10. anomaly detection applied to archaeology. IEEE – J. Sel. Top. Appl. 4, 870–876. 1061/(ASCE)0733-9453(2008)134:4(95). https://doi.org/10.1109/JSTARS.2011.2143696. Sony, 2017. α7R e-mount Camera With Full Frame Sensor. Full Specifications and Lucieer, A., Turner, D., King, D.H., Robinson, S.A., 2014. Using an Unmanned Aerial Features. URL: (Last date accessed: 1 Oct. 2018). https://www.sony.com/ Vehicle (UAV) to capture micro-topography of Antarctic moss beds. Int J Appl Earth electronics/interchangeable-lens-cameras/ilce-7r/specifications#specifications. Obs. 27, 53–62. https://doi.org/10.1016/j.jag.2013.05.011. Tian, J., Wang, L., Li, X., Gong, H., Shi, C., Zhong, R., Liu, X., 2017. Comparison of UAV Manfreda, S., McCabe, M., Miller, P., Lucas, R., Pajuelo Madrigal, V., Mallinis, G., Ben- and WorldView-2 imagery for mapping leaf area index of mangrove forest. Int. J. Dor, E., Helman, D., Estes, L., Ciraolo, G., et al., 2018. On the use of unmanned aerial Appl. Earth Obs. 61, 22–31. https://doi.org/10.1016/j.jag.2017.05.002. systems for environmental monitoring. Remote Sens. 10, 641–669. https://doi.org/ Turner, D., Lucieer, A., Malenovský, Z., King, D.H., Robinson, S.H., 2014a. Spatial Co- 10.3390/rs10040641. registration of ultra-high resolution visible, multispectral and thermal images ac- Martínez-Carricondo, P., Agüera-Vega, F., Carvajal-Ramírez, F., Mesas-Carrascosa, F.J., quired with a Micro-UAV over antarctic moss beds. Remote Sens. 6, 4003–4024. García-Ferrer, A., Pérez-Porras, F.J., 2018. Assessment of UAV-photogrammetric https://doi.org/10.3390/rs6054003. mapping accuracy based on variation of ground control points. Int. J. Appl. Earth Turner, D., Lucieer, A., Wallace, L., 2014b. Direct georeferencing of ultrahigh-resolution Obs. 72, 1–10. https://doi.org/10.1016/j.jag.2018.05.015. UAV imagery. IEEE Trans. Geosci. Remote 52, 2738–2745. https://doi.org/10.1109/ MicaSense, 2015. MicaSense RedEdge™ 3 Multispectral Camera User Manual. MicaSense, TGRS.2013.2265295. Inc., Seattle, WA, pp. 33. URL: (Last date accessed: 1 Oct. 2018). https://support. USGS-NUPO(a) (United States Geological Survey - National Unmanned Aircraft Systems micasense.com/hc/en-us/article_attachments/204648307/RedEdge_User_Manual_ Project Office), 2017. Unmanned Aircraft Systems Data Post-processing. Structure 06.pdf. From Motion Photogrammetry. Section – 1 Digital Single-lens Reflex (DSLR) Imagery. NASA (National Aeronautics and Space Administration), 2018. Landsat Data Continuity URL: (Last date Accessed: 1 Oct. 2018). https://uas.usgs.gov/pdf/ Mission (LCDM). URL: (Last date accessed: 1 Oct. 2018). https://www.nasa.gov/ PhotoScanProcessingDSLRMar2017.pdf. mission_pages/landsat/main/index.html. USGS-NUPO(b) (United States Geological Survey - National Unmanned Aircraft Systems Picard, R., Cook, D., 1984. Cross-validation of regression models. J. Am. Stat. Assoc. 79, Project Office), 2017. Unmanned Aircraft Systems Data Post-processing. Structure 575–583. https://doi.org/10.1080/01621459.1984.10478083. From Motion Photogrammetry. Section – 2 MicaSense 5-band Multispectral Imagery. PNOA (Plan Nacional de Ortofotografía Aérea), 2018. Productos generados” [Generated URL: (Last date Accessed: 1 Oct. 2018). https://uas.usgs.gov/pdf/ products]. URL: (Last date accessed: 1 Oct. 2018). http://pnoa.ign.es/productos. PhotoScanProcessingMicaSenseMar2017.pdf17.pdf. Pons, X., 2016. "MiraMon. Sistema d’Informació Geogràfica i software de Teledetecció. Voightländer, 2017. E-Mount 15 Mm / F 4.5 Super Wide Heliar Aspherical III. URL: (Last Versió 8.01b" [MiraMon. Geographical Information System and Remote Sensing date Accessed: 1 Oct. 2018). https://www.voigtlaender.de/lenses/e-mount/15-mm- software. Version 8.01b]. Centre de Recerca Ecològica i Aplicacions Forestals, 14-5-super-wide-heliar-aspherical-iii/?lang=en. CREAF. Bellaterra. ISBN: 84-931323-4-9. URL: http://www.creaf.uab.cat/miramon/ Whitehead, K., Hugenholtz, C., 2014. Remote sensing of the environment with small (Last date accessed: 1 Oct. 2018). unmanned aircraft Systems (UASs), part 1: a review of progress and challenges. J. Rehak, M., Skaloud, J., 2017. Time synchronization of consumer cameras on Micro aerial Unmanned Veh. Syst. 7, 69–85. https://doi.org/10.1139/juvs-2014-0007. Vehicles. ISPRS – J. Photogramm. 123, 114–123. https://doi.org/10.1016/j.isprsjprs. Wouda, F.J., Giuberti, M., Bellusci, G., Veltink, P.H., 2016. Estimation of full-body poses 2016.11.009. using only five inertial sensors: an eager or lazy learning approach? Sensors 16, Rehak, M., Mabillard, R., Skaloud, J., 2013. A Micro-UAV with the capability of direct 2138–2155. https://doi.org/10.3390/s16122138. georeferencing. ISPRS - International Archives of the Photogrammetry, Remote Zabala, S., 2017. Comparison of Multi-temporal and Multispectral Sentinel-2 and Sensing and Spatial Information Sciences, Volume XL-1/W2, 2013 UAV-g2013; 2013 Unmanned Aerial Vehicle Imagery for Crop Type Mapping”. Master of Science (MSc) Sep. 4–6; Rostock, Germany 317–323. https://doi.org/10.5194/isprsarchives-XL-1- Thesis in Geo-information Science and Earth Observation for Environmental W2-317-2013. Modelling and Management. June 2017. URL:. Lund University, Lund, Sweden. Rehak, M., Skaloud, J., 2015. Fixed-wing micro aerial vehicle for accurate corridor http://lup.lub.lu.se/luur/download?func=downloadFile&recordOId=8917610& mapping. ISPRS - Annals of the Photogrammetry, Remote Sensing and Spatial fileOId=8917627. 140