TY - JOUR
AU1 - Tomaštík, Julián
AU2 - Chudá, Juliána
AU3 - Tunák, Daniel
AU4 - Chudý, František
AU5 - Kardoš, Miroslav
AB - Abstract Smartphones with their capability to receive Global Navigation Satellite Systems (GNSS) signals can be currently considered the most common devices used for positioning tasks, including forestry applications. This study focuses on possible improvements related to two crucial changes implemented into Android smartphone positioning in the last 3 years – dual-frequency (L1/L5) GNSS receivers and the possibility of recording raw GNSS data. The study comprises three experiments: (1) real-time measurements of individual points, (2) real-time recording of trajectories, and (3) post-processing of raw GNSS data provided by the smartphone receiver. The real-time tests were conducted using final positions provided by the internal receiver, i.e. without further processing or averaging. The test on individual points has proven that the Xiaomi Mi8 smartphone with a multi-constellation, dual-frequency receiver was the only device whose accuracy was not significantly different from single-frequency mapping-grade receiver under any conditions. The horizontal accuracy of most devices was lower during leaf-on season (root mean square errors between 5.41 and 12.55 m) than during leaf-off season (4.10–11.44 m), and the accuracy was significantly better under open-area conditions (1.72–4.51 m) for all tested devices when compared with forest conditions. Results of the second experiment with track recording suggest that smartphone receivers are better suited for dynamic applications – the mean shift between reference and measured trajectories varied from 1.23 to 5.98 m under leaf-on conditions. Post-processing of the raw GNSS data in the third experiment brought very variable results. We achieved centimetre-level accuracy under open-area conditions; however, in forest, the accuracies varied from meters to tens of meters. Observed loss of the signal strength in the forest represented ~20 per cent of the open-area value. Overall, the multi-constellation, dual-frequency receiver provided more robust and accurate positional solutions compared with single-frequency smartphones. Applicability of the raw GNSS data must be further studied especially in forests, as the provided data are highly susceptible to multipath and other GNSS adverse effects. Introduction Spatial information about forests is needed for environmental monitoring and research, nature conservation, forest management and particularly for ‘precision forestry’ (Taylor et al., 2002; Holopainen et al., 2014). A current trend is to acquire as much spatial information about forests as possible using remote sensing techniques. This trend is driven mainly by the effectiveness of remote sensing approaches in combination with sufficient accuracy (Dash et al., 2016; White et al., 2016; Sačkov et al., 2019). However, despite the increased use of remote sensing data, terrestrial surveying methods are still required in many situations. In case of environmental research, these situations include collection of reference (ground truth) data (Fassnacht et al., 2017; Apostol et al., 2018), georeferencing of remote sensing data (Guerra-Hernández et al., 2017; Moudrý et al., 2019), placement of test plots (Murgaš et al., 2018), etc. When comparing location measurements in forested environments with measurements taken under conditions outside the forest, the measurements in forests are more complicated in most cases. In the ‘pre-GPS’ era, when the main positioning instruments were theodolite, compass and a measurement tape, this complication was mainly related to obstructed visibility and low density of target points, which could be used to reference the data into regional or global coordinate systems. With the advent of Global Navigation Satellite Systems (GNSS), these problems were solved, but others have arisen. Since the very beginning of GNSS research in forests, it was clear that complex forest structure adversely affects the signal reception from GNSS (Piedallu and Gégout, 2005; Pérez, 2006; Ordóñez Galán et al., 2011). These problems are mostly caused by blocking and multipath reflection of the GNSS signals by trees’ stems and leaves. Another obstruction can often come from rugged terrain. Despite the undoubted development of the technology and effort dedicated to solving these problems, forests and urban environments (with high number of tall and reflective objects) remain typical examples of areas with GNSS adverse conditions (McGaughey et al., 2017; Brach et al., 2019; Merry and Bettinger, 2019). Although the mentioned problems complicate the usage of GNSS measurements in forests and reduce their accuracy compared with open-area conditions, the application of these measurements was progressively adopted in forestry. The ‘close-to-real-time’ information about the position, independent from existing geodetic points, facilitated more comfortable navigation, easier location of management spots, the assessment of ownership boundaries, mapping of logged and disturbed areas as well as some other tasks, which were summarized in a recent survey (Bettinger et al., 2019). Besides these already well-established applications, more sophisticated solutions and combinations of GNSS use in forests emerge, including geofencing, activity recognition, Internet of Things, mesh networking and others (Keefe et al., 2019). With regard to instruments, low-cost receivers are the most frequently used by forestry practitioners (Bettinger et al., 2019). These include all recreational- and mapping-grade receivers without differential corrections, but smartphones are dominant. More expensive mapping- and survey-grade receivers are typically used only by dedicated specialists. The predominance of smartphones is logical not only in terms of their availability to users but also their flexibility and ability to solve tasks beyond positioning alone (Kennedy et al., 2014; Vastaranta et al., 2015; Bianchi et al., 2017; Marzulli et al., 2020). In contrast to the high adoption rate, comprehensive studies on the positional accuracy of smartphones are sparse, with some exception of studies focusing on raw GNSS data evaluations. Under open-area conditions, Schaefer and Woodyer (2015) reported absolute accuracies between 2.64 and 4.19 m for selected Apple, Sony and Samsung devices. Szot et al. (2019) tested the Samsung Galaxy series and obtained root mean square errors between 1.02 and 6.84 m using a 24-h observation period. Bauer (2013) noticed that the smartphone accuracy can be influenced by the positioning application. In an urban environment, Merry and Bettinger et al. (2019) studied the influence of tree vegetation on the measurements taken by an Iphone 6 and reported average horizontal errors of 7–13 m. In our previous study (Tomaštík et al., 2017), we obtained accuracies of 6.74–11.45 m (leaf-on) and 4.51–6.72 m (leaf-off) using three smartphones under forest conditions. GNSS receivers in smartphones typically belong to the family of ‘high-sensitivity’ receivers (Schwieger, 2009; Zandbergen and Barbeau, 2011), which are capable of receiving GNSS signals with a power ratio below −150 dBm. The sensitivity of ‘normal’ receivers is around −130 dBm, thus restricting the reception of weaker signals and indoor usage. However, the higher sensitivity cannot be confused with higher positional accuracy. Another functionality of smartphones is the so-called ‘assisted GPS’ (A-GPS) (van Diggelen, 2009). This feature uses the ability of smartphones to communicate with the Internet to derive information about satellite signals, which should be available in the approximate location of the smartphone. Thus, the receiver does not have to search through all possible signals and the Time to First Fix can be reduced (Zandbergen and Barbeau, 2011). Most mid- and high-tier smartphone GNSS receivers can receive signals from multiple satellite systems. There are receivers capable of receiving signals from all systems operational at a global scale including NAVSTAR GPS (USA), GLONASS (Russia), Galileo (EU), Beidou (China) as well as regional systems such as Japan’s QZSS or India’s IRNSS. Higher numbers of usable satellites provide better positioning services. In 2018, Xiaomi introduced the first smartphone with a dual-frequency GNSS receiver. The incorporated Broadcom BCM47755 receiver can receive the L5 (GPS) and E5 (Galileo) signals in addition to L1 signals available for single-frequency receivers. Thus, the number of available signals increases even more. The L5 signals are more advanced in comparison with the ‘legacy’ L1 signals and were designed to mitigate some problems, which arose during the initial usage of the L1. From a forester’s point of view, their higher robustness against the multipath effect (Irsigler et al., 2004) can be beneficial. Combining L1 and L5 signals can also improve measurement accuracy via ionospheric corrections (Fortunato et al., 2019). Another important change in Android smartphone positioning was announced by Google in 2016. Until then, the smartphone GNSS receivers worked as ‘black boxes’, providing completely processed positions with very little possibilities of influencing or modifying the process. The ability to record raw GNSS measurements was introduced with the application programming interface (API) 24 (GSA Raw Measurements Task Force, 2017). In theory, every smartphone with Android 7.0 or higher should be capable of recording pseudorange, carrier-phase, Doppler and signal-to-noise ratio (SNR) observables (Fortunato et al., 2019). However, this must also be enabled by hardware. A list of devices supporting full or partial raw GNSS measurements is maintained by Google (Google, 2019). In principle, the raw GNSS data from smartphones can be subsequently processed like the data from higher grade receivers using various sources of corrections, e.g. another receiver or a continuously operating reference station (CORS) (Lachapelle et al., 2018; Dabove and Di Pietra, 2019; Paziewski et al., 2019; Robustelli et al., 2019), precise ephemeris of satellites (Romero-Andrade et al., 2019; Wu et al., 2019), etc. Even some automated processing services are being adapted to process the raw GNSS data provided by low-cost receivers, which are characterized by lower signal quality (Banville et al., 2019). The mentioned studies were focused mainly on the performance of raw GNSS measurements under optimal conditions for signal reception, while some of them described initial problems in urban environment with a higher multipath effect. However, we are not aware of any studies addressing this matter in forests. The main aim of this study was to evaluate possible benefits arising from the new availability of multi-constellation, dual-frequency receivers in smartphones as well as an evaluation of potential benefits of raw GNSS data acquired using such a receiver for positioning in forests. Sub-aims included a comparison of the accuracies of single- and dual-frequency smartphones for real-time measurement of individual points and trajectories in a forest, and an evaluation of static GNSS measurements with the use of raw GNSS data collected by a smartphone under forest conditions. Methods The study is divided into three experiments. The first experiment examined the accuracy achievable for real-time measurements of individual points. In the second experiment, we evaluated the accuracy of trajectories recorded using smartphone GNSS. The positions in these two experiments were provided directly by the internal GNSS receivers and were not corrected in any way. The last experiment evaluated the option to post-process the raw GNSS data acquired using the GNSS receiver included in a smartphone. The priority in all experiments was the evaluation of the internal GNSS receivers’ capabilities; however, in the first and second experiment, we also compare the performance of multiple internal smartphone receivers to the performance of an external Bluetooth receiver and a mapping-grade receiver. A list of the devices is summarised in Table 1. Although the Trimble Nomad 5 is a mapping grade receiver, it also uses the Android OS. An external Bluetooth receiver GNS2000 was connected to the ZTE Blade smartphone, which served only to display and record the data. Table 1 Devices used in the experiments and their basic characteristics. Device . Type . Supported GNSS constellations (and frequencies)1 . LG G2 Smartphone GPS(L1), GLO(L1) Lenovo A5000 Smartphone GPS(L1) Lenovo Phab 2 Pro Smartphone GPS(L1), GLO(L1), Bei(B1) Huawei P20 lite Smartphone GPS(L1), GLO(L1), Bei(B1) Xiaomi Mi8 Smartphone GPS(L1 + L5), GLO(L1), Gal(E1 + E5), Bei(B1), QZS(L1 + L5) GNS2000 Bluetooth receiver2 GPS(L1), GLO(L1) Trimble Nomad 5 Handheld GIS receiver GPS(L1), GLO(L1), Gal(E1), Bei(B1) Device . Type . Supported GNSS constellations (and frequencies)1 . LG G2 Smartphone GPS(L1), GLO(L1) Lenovo A5000 Smartphone GPS(L1) Lenovo Phab 2 Pro Smartphone GPS(L1), GLO(L1), Bei(B1) Huawei P20 lite Smartphone GPS(L1), GLO(L1), Bei(B1) Xiaomi Mi8 Smartphone GPS(L1 + L5), GLO(L1), Gal(E1 + E5), Bei(B1), QZS(L1 + L5) GNS2000 Bluetooth receiver2 GPS(L1), GLO(L1) Trimble Nomad 5 Handheld GIS receiver GPS(L1), GLO(L1), Gal(E1), Bei(B1) 1GLO – GLONASS, Gal – Galileo, Bei – Beidou, QZS – Quasi-Zenith Satellite System. 2Connected to ZTE Blade smartphone. Open in new tab Table 1 Devices used in the experiments and their basic characteristics. Device . Type . Supported GNSS constellations (and frequencies)1 . LG G2 Smartphone GPS(L1), GLO(L1) Lenovo A5000 Smartphone GPS(L1) Lenovo Phab 2 Pro Smartphone GPS(L1), GLO(L1), Bei(B1) Huawei P20 lite Smartphone GPS(L1), GLO(L1), Bei(B1) Xiaomi Mi8 Smartphone GPS(L1 + L5), GLO(L1), Gal(E1 + E5), Bei(B1), QZS(L1 + L5) GNS2000 Bluetooth receiver2 GPS(L1), GLO(L1) Trimble Nomad 5 Handheld GIS receiver GPS(L1), GLO(L1), Gal(E1), Bei(B1) Device . Type . Supported GNSS constellations (and frequencies)1 . LG G2 Smartphone GPS(L1), GLO(L1) Lenovo A5000 Smartphone GPS(L1) Lenovo Phab 2 Pro Smartphone GPS(L1), GLO(L1), Bei(B1) Huawei P20 lite Smartphone GPS(L1), GLO(L1), Bei(B1) Xiaomi Mi8 Smartphone GPS(L1 + L5), GLO(L1), Gal(E1 + E5), Bei(B1), QZS(L1 + L5) GNS2000 Bluetooth receiver2 GPS(L1), GLO(L1) Trimble Nomad 5 Handheld GIS receiver GPS(L1), GLO(L1), Gal(E1), Bei(B1) 1GLO – GLONASS, Gal – Galileo, Bei – Beidou, QZS – Quasi-Zenith Satellite System. 2Connected to ZTE Blade smartphone. Open in new tab Experiment 1: point measurement accuracy The first experiment was conducted by measuring a set of points located near the village Sielnica in Slovakia (~48°38′ N, 19°05′ E). This set consisted of 43 points in a forest and 15 points in an open area (Figure 1). Figure 1 Open in new tabDownload slide Experimental area with test points and trajectories, and its location within Slovakia. This figure appears in colour in the online version of Forestry. Figure 1 Open in new tabDownload slide Experimental area with test points and trajectories, and its location within Slovakia. This figure appears in colour in the online version of Forestry. The tree species composition in the area is ~30 per cent sessile oak (Quercus petraea Matt.), 30 per cent European beech (Fagus sylvatica L.), 25 per cent Norway spruce (Picea abies L.) and 15 per cent other species (hornbeam – Carpinus betulus L., pine – Pinus sylvestris L., maple – Acer sp., lime tree – Tilia cordata Mill., etc.) It was possible to identify several groups of points representing various stand conditions. Forest stand characteristics for these groups are in Table 2 and examples of stand conditions are shown in Figure 2. Apart from the characteristics given in the table, the mature mixed and mature coniferous stands are characterized by almost fully closed forest canopies. The mature coniferous stand is only 20-m wide and ~90-m long, enclosed by a mixed forest. The inner edge of the mature mixed stand is characterized by the absence of tree stems in the direction to the open area. This part of the sky-view is covered only by branches. The post-disturbance opening is 20-m wide and 110-m long. The road crossing the test area is 3-m wide; however, trees (stems) in the young forest are significantly closer to the road compared with the mature forest. These stand characteristics were used to assess the influence of various forest settings on the resulting accuracy of GNSS measurements. However, as the number of points representing differing stand conditions is predictably too sparse for a robust statistical analysis, we mainly focus on the overall results for the forest. Points located on the transition between the particular forest stand conditions were not included in the aforementioned groups, but they were included in the overall results for the forest. Points in the open area, numbered from 100 to 114 (Figure 1), were placed along telephone poles on a meadow with minimal number of obstacles degrading the GNSS signals. The poles were used to facilitate the orientation on the meadow. The distance between a measured point and a pole was 1.5 m to limit the influence of the pole on received GNSS signals. The points were always to the North–West from the imaginary line connecting points from No. 100 to No. 114. Overall, the locations of our points are similar to the locations used in the previous study (Tomaštík et al., 2017), except for few areas, which were changed due to ongoing logging operations. Table 2 Basic characteristics of specific forest stand conditions in the experimental area. Conditions . Forest stand characteristics1 . Open sky2 . Included points (No.) . Marking3 . Age . Mean height (m) . Mean DBH (cm) . Volume per ha (m3) . Mature mixed forest 75 25 30 390 10% 13, 14, 15, 16, 17 (a) Mature coniferous forest 60 26 31 518 10% 25, 26, 27, 28 (b) Post-disturbance opening Located between stands (a) and (e) 50% 9, 10, 11, 12 (c) Inner edge of mature forest 75 25 30 390 15% 19, 20, 21, 22, 23, 24 (d) Road in young forest 35 17 17 188 20% 29, 30, 31, 32, 33 (e) Road in mature forest 75 25 30 390 30% 2, 3, 4, 5, 6, 7 (f) Conditions . Forest stand characteristics1 . Open sky2 . Included points (No.) . Marking3 . Age . Mean height (m) . Mean DBH (cm) . Volume per ha (m3) . Mature mixed forest 75 25 30 390 10% 13, 14, 15, 16, 17 (a) Mature coniferous forest 60 26 31 518 10% 25, 26, 27, 28 (b) Post-disturbance opening Located between stands (a) and (e) 50% 9, 10, 11, 12 (c) Inner edge of mature forest 75 25 30 390 15% 19, 20, 21, 22, 23, 24 (d) Road in young forest 35 17 17 188 20% 29, 30, 31, 32, 33 (e) Road in mature forest 75 25 30 390 30% 2, 3, 4, 5, 6, 7 (f) 1Based on a forest management plan. 2This measure was determined under leaf-on conditions. 3According to Figure 2. Open in new tab Table 2 Basic characteristics of specific forest stand conditions in the experimental area. Conditions . Forest stand characteristics1 . Open sky2 . Included points (No.) . Marking3 . Age . Mean height (m) . Mean DBH (cm) . Volume per ha (m3) . Mature mixed forest 75 25 30 390 10% 13, 14, 15, 16, 17 (a) Mature coniferous forest 60 26 31 518 10% 25, 26, 27, 28 (b) Post-disturbance opening Located between stands (a) and (e) 50% 9, 10, 11, 12 (c) Inner edge of mature forest 75 25 30 390 15% 19, 20, 21, 22, 23, 24 (d) Road in young forest 35 17 17 188 20% 29, 30, 31, 32, 33 (e) Road in mature forest 75 25 30 390 30% 2, 3, 4, 5, 6, 7 (f) Conditions . Forest stand characteristics1 . Open sky2 . Included points (No.) . Marking3 . Age . Mean height (m) . Mean DBH (cm) . Volume per ha (m3) . Mature mixed forest 75 25 30 390 10% 13, 14, 15, 16, 17 (a) Mature coniferous forest 60 26 31 518 10% 25, 26, 27, 28 (b) Post-disturbance opening Located between stands (a) and (e) 50% 9, 10, 11, 12 (c) Inner edge of mature forest 75 25 30 390 15% 19, 20, 21, 22, 23, 24 (d) Road in young forest 35 17 17 188 20% 29, 30, 31, 32, 33 (e) Road in mature forest 75 25 30 390 30% 2, 3, 4, 5, 6, 7 (f) 1Based on a forest management plan. 2This measure was determined under leaf-on conditions. 3According to Figure 2. Open in new tab Figure 2 Open in new tabDownload slide Examples of forest stand conditions present in the experimental area: (a) mature mixed forest, (b) mature coniferous forest, (c) post-disturbance opening, (d) inner edge of mature forest, (e) road in young forest and (f) road in mature forest. This figure appears in colour in the online version of Forestry. Figure 2 Open in new tabDownload slide Examples of forest stand conditions present in the experimental area: (a) mature mixed forest, (b) mature coniferous forest, (c) post-disturbance opening, (d) inner edge of mature forest, (e) road in young forest and (f) road in mature forest. This figure appears in colour in the online version of Forestry. Prior to reference measurements, all 58 points were marked using wood or iron stakes (in the forest and on the meadow) or using a colour cross with a nail in the middle (on the road). The reference positions of the points were measured with a Topcon GPT-9000M total station using a combination of polygonal traverses and additional measurements using polar coordinates (angle and distance). This ‘traditional’ surveying method, which is capable of measuring distances with millimetre accuracy and angles with an accuracy of a few seconds, provided GNSS independent reference measurements. Positional accuracies of these measurements, taking also the measurement method into the account, were less than 5 cm. The national coordinate reference system S-JTSK (EPSG:5514) was used to collect the reference coordinates. This system, designed for the Czech Republic and Slovakia, is an oblique case of the Lambert conformal conic projection and uses Cartesian coordinates in meters. In total, we examined five smartphones and two other GNSS receivers (Table 1). For better manageability, most of the devices were attached to a supporting construction (Figure 3). The centre of the supporting construction was placed over a measured point. As the exact position of smartphones’ phase centres was not known, we used the centre of the device as a reference point. The final position of the device was subsequently edited using distance and direction of this reference point from the centre of the supporting construction, with regard to the trajectory bearing. However, these shifts were lower than 20 cm. Two other devices (Huawei P20 lite and Trimble Nomad 5) were operated by a second operator closely following the first operator. These devices were held in hand. The trajectory of both operators was the same, the distance between them was equal to the distance of two adjacent measured points (i.e. operator 1 on point ‘n’, operator 2 on point ‘n – 1’). Thus, all measurements for one point were completed within a 3-min time period. Figure 3 Open in new tabDownload slide Experimental setup used to operate the tested devices in the first and second experiments. The INS (in backpack) was used only for the second experiment. This figure appears in colour in the online version of Forestry. Figure 3 Open in new tabDownload slide Experimental setup used to operate the tested devices in the first and second experiments. The INS (in backpack) was used only for the second experiment. This figure appears in colour in the online version of Forestry. The measurements were conducted under leaf-off (29 March 2019) and leaf-on conditions (21 June 2019). The Locus Map application (Asamm Software, Prague, CZ) was used to record points’ positions in all tested devices. To maintain the uninterrupted functionality of GNSS receivers, we enabled the ‘Track recording’ function. Subsequently, the positions of individual points were recorded using the ‘Add new point’ function. A single position was recorded for every point, with no averaging. Positions (WGS84 coordinates) were exported as text files and transformed into the S-JTSK system using the official transformation service (SKPOS, 2019). Root mean square horizontal errors (RMSExy) were used as a basic measure to compare devices and conditions. As a common measure in accuracy evaluations, RMSEs also allow us to compare this method with other surveying methods. As the first step, root mean square coordinate errors were calculated as follows: $$\begin{equation} {\mathrm{RMSE}}_x=\sqrt{\frac{\sum_{i=1}^n\Delta{x_i}^2}{n}} \end{equation}$$(1) $$\begin{equation} {\mathrm{RMSE}}_y=\sqrt{\frac{\sum_{i=1}^n\Delta{y_i}^2}{n}}, \end{equation}$$(2) where Δxi and Δyi are differences between reference coordinates and coordinates acquired using GNSS. Subsequently, the RMSExy was calculated $$\begin{equation} {\mathrm{RMSE}}_{xy}=\sqrt{{{\mathrm{RMSE}}_x}^2+{{\mathrm{RMSE}}_y}^2}. \end{equation}$$(3) To compare pairwise differences between leaf-off, leaf-on and open-area conditions, we employed a modification of the F-test, called ‘coefficient of relative efficiency’ (Šmelko, 2007) $$\begin{equation} {R}_e=\frac{{{\mathrm{RMSE}}_A}^2}{{{\mathrm{RMSE}}_B}^2}, \end{equation}$$(4) where RMSEA is the mean error with a higher value and RMSEB is the mean error with a lower value. Statistical significance of the differences between these means is assessed when comparing the Re value to the upper critical value of the F-distribution. The critical value for comparison between leaf-on and leaf-off conditions was F(0.05; 42; 42) = 1.67097, while for leaf-off vs open-area conditions, the value was F(0.05; 42; 29) = 1.79832. RMSEs provide poor information about the within-group variability and are prone to outliers and extremes, which are common for GNSS measurements conducted using low-cost receivers. Therefore, to facilitate the variability analysis, positional errors of individual points were calculated as follows: $$\begin{equation} {\Delta p}_i=\sqrt{{{\Delta x}_i}^2+{{\Delta y}_i}^2}, \end{equation}$$(5) Minima, maxima, averages and standard deviations (SD) were used to compare the error of the tested devices. Subsequently, we conducted a one-way ANOVA with Tukey HSD post hoc test to evaluate the significance of the between-device differences using the STATISTICA software (Statsoft, Inc., Tulsa, USA). Two detailed analyses were conducted only for selected devices – the Lenovo A5000 (GPS L1 only) and Xiaomi Mi 8 (multi-constellation, dual-frequency) smartphones, and the mapping-grade receiver Trimble Nomad 5. In the first case, the influence of differing forest stand conditions (see Table 2) was evaluated using means, SDs, maxima and minima of positional errors obtained in the particular groups of points. The second analysis focused on the number of available satellite signals and a presumably significant difference between the older and the newer generation of smartphones. The number of satellite signals was recorded using National Marine Electronics Association messages on the epoch basis (every second) during the measurements. Median values were used to compare the devices and conditions. A Kruskal–Wallis test in the STATISTICA software was used to evaluate significance of differences between the devices. Figure 4 Open in new tabDownload slide Assessment of trajectories’ accuracy using buffers around the reference trajectory. Example of trajectory No. 1 (all trajectories available as Supplementary Figures 1–3). This figure appears in colour in the online version of Forestry. Figure 4 Open in new tabDownload slide Assessment of trajectories’ accuracy using buffers around the reference trajectory. Example of trajectory No. 1 (all trajectories available as Supplementary Figures 1–3). This figure appears in colour in the online version of Forestry. Experiment 2: trajectory accuracy Three trajectories, enclosing differing parts of the forest (Figure 1), were used to evaluate the applicability and accuracy of the smartphones and two GNSS receivers for track recording with the exception of the Huawei P20 lite smartphone, which was excluded due to technical problems. The measurements were conducted under leaf-on conditions (10 August 2019). Reference trajectories were acquired using a pedestrian inertial navigation system (INS). This system consisted of the Novatel ProPak 6 GNSS receiver with a KVH-1750 inertial measurement unit enclosed in a rugged backpack (Figure 3). Previous test of this INS showed that the accuracy of a trajectory collected with this device is within 1 m even under forest canopy (Tuček et al., 2016). In case of smartphones and handheld receiver, the setup was the same as in the previous experiment. Trajectories of all devices were recorded simultaneously, including the INS. In this experiment, the only device operated by the second operator was the Trimble Nomad 5. The trajectories of both operators were the same; the distance between them was ~2 m. All mobile devices were held at the height of 1.2–1.4 m above ground, which is common when operating such devices. The trajectories represented boundaries of forest management units, which were marked using 20 × 5 cm white signs on trees, according to the Slovak legislation. However, with regard to INS requirements, the start and end point as well as a short straight segment of each trajectory (identifiable as the overlapping sections on the roadways on Figure 4 and Supplementary Figures 1–3) were situated under conditions without any obstacles influencing the GNSS signal reception. These were used to achieve proper static and kinematic alignment of the INS. The trajectories were recorded using fluent movement without longer pauses. Trajectories from the Android devices were recorded using the Locus Map application and ‘Track recording’ function. For the INS and all mobile devices, new points were added to the trajectories at a 1-Hz rate. Thus, every segment (or even single point) recorded by the Android devices has its own reference. Trajectories of the INS were processed in the Inertial Explorer software (Novatel, Inc.) as a combined, tightly coupled GNSS+IMU solution. Basic statistical measures of the estimated errors for all trajectories are reported as the first part of this experiment’s results. Tracks acquired using the Android devices were exported in the GPS Exchange Format (*.gpx) and subsequently processed in the QGIS 3.8.3 software (open source software (QGIS Development Team, 2019)). Buffers from 1 to 10 m (with steps of 1 m) were created around the reference INS trajectories (Figure 4). Trajectories from Android devices were intersected with these buffers; percentual shares of the trajectories belonging to individual buffers were calculated with regard to the overall length of particular trajectory. Thus, the evaluation was based on linear segments belonging to individual buffers, not individual points. Parts of the trajectories, which were out of the 10-m buffer, were classified as the 10+ meter category. This 10-m limit was chosen after initial evaluation, and values above this limit were considered coarse errors. However, to calculate the average shift, the 10+ category was further intersected with buffers of up to 40 m, based on the highest errors. This was not possible to conduct in one step, because the 40-m buffers were too wide and caused many intersections, especially in corners of the trajectories. The average shift of a trajectory was calculated as a mean value using percentual shares of the trajectory segments belonging to individual buffers and mean distance values of the buffers (i.e. 0.5 m for buffer 0–1 m, 1.5 m for buffer 1–2 m, etc.). These errors were calculated for individual trajectories and overall. Acreages enclosed by the trajectories and tracks’ lengths were also determined and compared with the reference. Differences from reference lengths and areas were calculated relatively (as percentual shares) from reference values. Experiment 3: raw measurements post-processing accuracy The Xiaomi Mi 8 was the only tested device capable of providing raw GNSS measurements. This test was conducted under leaf-on conditions (20 August 2019) on four points in the forest, which represented a mature mixed stand (point No. 15), mature coniferous stand (No. 26), a post-disturbance opening (No. 11) and an inner edge of a mature forest stand (No. 22) (Figure 2a–d) as well as one point under open-area conditions (No. 103). The reference coordinates of these points were the same as in Experiment 1, obtained using the Topcon GPT-9000M total station. The smartphone was placed on a 1.45-m high pole with tripod, which provided stabilization and levelling of the device (Figure 5). The device was placed on a ‘ground plane’ consisting of a 2-mm-thick solid steel plate, exceeding the phone dimensions by 2–3 cm. This ‘ground plane’ was used to eliminate signals reflected from the ground. The longest axis of the phone was oriented towards the North–South direction using a compass. Figure 5 Open in new tabDownload slide Setup used for static recording of raw GNSS measurements. This figure appears in colour in the online version of Forestry. Figure 5 Open in new tabDownload slide Setup used for static recording of raw GNSS measurements. This figure appears in colour in the online version of Forestry. Figure 6 Open in new tabDownload slide Minima, maxima, SDs and means of positional errors according to tested device and conditions. This figure appears in colour in the online version of Forestry. Figure 6 Open in new tabDownload slide Minima, maxima, SDs and means of positional errors according to tested device and conditions. This figure appears in colour in the online version of Forestry. The Geo++ Rinex Logger application (Geo++ GmbH, Garbsen, Germany) was used to record raw GNSS measurements. The application records the data in the Rinex format, which can be subsequently processed by most GNSS processing software. We used a 1-h observation period; the data were recorded at 1-Hz rate. The ‘Force full GNSS Measurements’ option was turned on under ‘Developer option’ to avoid duty cycling. The data (both rover and base) are available at Mendeley Data service (Tomaštík, 2019). The acquired Rinex files were post-processed in the Demo5 modification of the open-source RTKLib software, which is currently developed by Tim Everett (Everett, 2019). This modification is developed with special emphasis on developing routines suitable to process data provided by low-cost receivers. The RTKPOST executable was used for processing the raw GNSS data with the following settings (available in ‘Options…’): Positioning mode: static Frequencies: L1 and L1 + L2 + E5b + L5 Filter type: combined Elevation mask: 10 Constellations: GPS + GLONASS and GPS + GLONASS + Galileo + QZSS + Beidou Integer ambiguity resolution: ‘Fix and hold’ and ‘Instantaneous.’ These settings were chosen to evaluate the influence of the second frequency (E5, L5) as well as newer satellite systems (Galileo, QZSS, Beidou). Basic difference between ‘Fix and hold’ and ‘Instantaneous’ settings for the ambiguity resolution is that while ‘Fix and hold’ uses a Kalman filter to predict changes between individual solution, ‘Instantaneous’ estimates and resolves integer ambiguities on epoch-by-epoch basis (Takasu, 2013). A CORS in Banská Bystrica was used as a base station. The base line to the reference station was ~13.5 km. Due to unspecified problems with Galileo navigation data provided by the CORS, we used navigation data provided by NASA (NASA, 2019). The average positions were calculated from ~3600 epochs collected during the 1-h measurement period. After calculation, solutions were plotted using the RTKPLOT executable, which can compare the average position to the reference and provide basic statistic characteristics of errors. Horizontal and vertical errors of the average positions were used as basic measures for comparison. The horizontal error was calculated from its easting and northing components provided by the software, similarly to formula (5) $$\begin{equation} {\Delta EN}_i=\sqrt{{{\Delta E}_i}^2+{{\Delta N}_i}^2}, \end{equation}$$(6) where ΔEi and ΔNi are differences between the reference and calculated coordinates of the average position. Vertical errors ΔHi were calculated as simple differences between reference and measured coordinates. In addition to the metrics available without the raw data (e.g. number and geometry of satellites), the raw GNSS data bring new possibilities to analyze GNSS data provided by smartphone GNSS receivers. One of the most common measures describing the signal strength is the carrier-to-noise density C/N0 (often mistaken with SNR, see (Joseph, 2010)). The C/N0 values for every epoch and satellite signal were exported using the RTKPLOT executable. Minima, maxima, means and SDs were computed for every available signal and all signals together. The main aim was to evaluate differences in signal strength between open-area and forest conditions. Finally, the number of recorded satellite signals’ observations was calculated and compared with regard to the measurement’s conditions and particular signals. Results Experiment 1 Point measurement accuracy Root mean square horizontal errors (RMSExy) according to tested devices and vegetation conditions are reported in Table 3. The RMSEs of all devices tested in the forest are lower under the leaf-off conditions, with exception of the Huawei P20 lite, where the error increases from 8.12 to 11.44 m. The pairwise test using the coefficient of relative efficiency confirms that differences between mean errors are significant (at α = 0.05) for GNS2000, Xiaomi Mi8, LG G2 and Lenovo A5000, but insignificant for Trimble Nomad 5 and Lenovo Phab 2 Pro. The difference is surprisingly also significant for Huawei P20 lite, but in the opposite way – the accuracy under leaf-on conditions was higher compared with the leaf-off conditions. When comparing the accuracies obtained in the forest and under open-area conditions, the mean errors for all devices are significantly lower for the open area compared with both leaf-on and leaf-off conditions. Table 3 Root mean square horizontal errors RMSExy (in meters) according to examined devices and conditions. Device . Conditions . Leaf-on . Leaf-off . Open area . Trimble Nomad 5 5.41 4.73 1.72 GNS2000 10.32 6.09 2.61 Xiaomi Mi 8 6.13 4.10 2.23 Huawei P20 lite 8.12 11.44 3.44 LG G2 12.55 8.84 2.09 Lenovo A5000 11.32 7.97 4.40 Lenovo Phab 2 Pro 8.43 8.36 4.51 Device . Conditions . Leaf-on . Leaf-off . Open area . Trimble Nomad 5 5.41 4.73 1.72 GNS2000 10.32 6.09 2.61 Xiaomi Mi 8 6.13 4.10 2.23 Huawei P20 lite 8.12 11.44 3.44 LG G2 12.55 8.84 2.09 Lenovo A5000 11.32 7.97 4.40 Lenovo Phab 2 Pro 8.43 8.36 4.51 Open in new tab Table 3 Root mean square horizontal errors RMSExy (in meters) according to examined devices and conditions. Device . Conditions . Leaf-on . Leaf-off . Open area . Trimble Nomad 5 5.41 4.73 1.72 GNS2000 10.32 6.09 2.61 Xiaomi Mi 8 6.13 4.10 2.23 Huawei P20 lite 8.12 11.44 3.44 LG G2 12.55 8.84 2.09 Lenovo A5000 11.32 7.97 4.40 Lenovo Phab 2 Pro 8.43 8.36 4.51 Device . Conditions . Leaf-on . Leaf-off . Open area . Trimble Nomad 5 5.41 4.73 1.72 GNS2000 10.32 6.09 2.61 Xiaomi Mi 8 6.13 4.10 2.23 Huawei P20 lite 8.12 11.44 3.44 LG G2 12.55 8.84 2.09 Lenovo A5000 11.32 7.97 4.40 Lenovo Phab 2 Pro 8.43 8.36 4.51 Open in new tab As the RMSEs give little information about the within-group variability and are prone to outliers and extremes, we used the positional errors calculated for every point to describe the variability. Average positional errors, their SDs and minima and maxima according to the devices and conditions are reported in Figure 6. The differences between devices were tested using ANOVA and a subsequent Tukey HSD test (results available as Supplementary Tables 1–3). The maxima and SDs of errors in most cases decrease, when we compare leaf-on, leaf-off and open-area conditions. Under the leaf-on conditions, the lowest mean positional error is observed for the Trimble Nomad 5 (4.55 m), while the highest was observed using the LG G2 (10.40 m). Mean errors of the smartphones Xiaomi Mi 8, Huawei P20 lite and Lenovo Phab 2 Pro are not significantly different from the mean error of Trimble Nomad 5 handheld GIS receiver. Under the leaf-off conditions, the lowest error (3.71 m) is observed for the Xiaomi Mi 8 and the highest (9.37 m) for the Huawei P20 lite. The mean errors of Trimble Nomad 5 and GNS2000 are not significantly different from the error of the Xiaomi Mi 8. The Trimble Nomad 5 achieved the best results also under open-area conditions (mean error of 1.64 m), while Lenovo Phab 2 Pro with an error of 4.04 m performed worst under these conditions. Results for GNS2000, Xiaomi Mi 8 and LG G2 are not significantly different from the results of the Trimble Nomad 5. We have no sound explanation of these very good results of the LG G2 under open-area conditions, because in the forest this device was one of the worst performing. Overall, the best results were achieved by Trimble Nomad 5 and Xiaomi Mi 8. Moreover, the Xiaomi Mi 8 with the dual-frequency receiver is the only device whose results are not significantly different from the results of Trimble Nomad 5 mapping-grade receiver considering all tested conditions. Forest stand settings influence The influence of differing forest stand settings was evaluated for selected devices – Lenovo A5000, Xiaomi Mi 8 and Trimble Nomad 5. Means, minima, maxima and SDs of positional errors according to conditions and devices are shown in Figure 7. Figure 7 Open in new tabDownload slide Minima, maxima, SDs, means and trend functions of positional errors according to tested device and various forest stand settings. This figure appears in colour in the online version of Forestry. Figure 7 Open in new tabDownload slide Minima, maxima, SDs, means and trend functions of positional errors according to tested device and various forest stand settings. This figure appears in colour in the online version of Forestry. Figure 8 Open in new tabDownload slide Number of available satellite signals during the leaf-on season measurement according to tested devices. Periods related to various environment settings are marked by letters: (a) road in mature forest, (b) post-disturbance opening, (c) mature mixed forest, (d) inner edge of mature forest, (e) mature coniferous stand, (f) road in young forest and (g) open area. This figure appears in colour in the online version of Forestry. Figure 8 Open in new tabDownload slide Number of available satellite signals during the leaf-on season measurement according to tested devices. Periods related to various environment settings are marked by letters: (a) road in mature forest, (b) post-disturbance opening, (c) mature mixed forest, (d) inner edge of mature forest, (e) mature coniferous stand, (f) road in young forest and (g) open area. This figure appears in colour in the online version of Forestry. The conditions have been arranged according to the open sky estimation, from the least to the most open. The mature coniferous stand was considered more adverse to GNSS measurements according to higher volume per hectare compared with the mature mixed stand. Despite this arrangement, hardly any common trend of the mean errors is visible. The only setting where the mean positional errors almost regularly decrease with increasing openness of the forest is the measurement using Trimble Nomad 5 under the leaf-off conditions. Considering other devices under leaf-off conditions, the trend function is decreasing for Lenovo A5000, but with a very slight slope, and it is constant for the Xiaomi Mi 8. Under leaf-on conditions, the mean errors increase with the increased openness for Xiaomi Mi 8 and Lenovo A5000. The accuracy increases with the openness for the Trimble Nomad 5; however, the slope of the trend is much lower than under leaf-off conditions. When the numeric values of the open sky ratio are taken into the account, only two correlations are significant at α = 0.05 – correlation for the Trimble Nomad 5 under leaf-off conditions (R = −0.498, P = 0.005) and for Xiaomi Mi 8 under leaf-on conditions, however, with a positive R of 0.549 (P = 0.002) in this case. Number of satellite signals The main aim of this part of the experiment was to describe changes between an older, single-constellation, single-frequency receiver and a multi-constellation, dual-frequency receiver with regard to the number of available satellite signals. The behavior of this metric for Lenovo A5000 (GPS L1 only), Xiaomi Mi 8 (multi-constellation, dual-frequency) and Trimble Nomad 5 (multi-constellation, single-frequency, mapping-grade receiver) during the measurements under leaf-on conditions is shown in Figure 8. With medians of 8 for Lenovo A5000 and 25 for Xiaomi Mi 8, the difference is significant. The maximum number of available signals is 9 for the older smartphone, while the maximum is 34 for the newer one. The third device in this evaluation, Trimble Nomad 5, shows an uncommon behavior with constant number of satellite signals (12) throughout the measurement. When considering the differing forest stand settings (Figure 8, parts a–f), the medians change between 24 and 26 for the Xiaomi Mi 8, and from 7 to 9 for Lenovo A5000. This change is significant for the older smartphone but cannot be credited to the changes of the forest settings alone, e.g. changes in the satellite constellation (geometry) must be taken into the account. For Xiaomi Mi 8, there is a visible increase in the number of available signals after the transition to the open area (Figure 8, part g, median = 29). During the leaf-off season, the median number of satellite signals was 32 for Xiaomi Mi 8, with a maximum of 37 signals. The difference between the forest and open-area conditions was unidentifiable. For Trimble Nomad 5, the median number of signals remained at 12 after a short period of initialization (30 s) during which the number varied between 9 and 12. For unknown reasons, Lenovo A5000 failed to record this measure during the leaf-off measurement. Table 4 Basic statistical and geometrical characteristics of the reference trajectories acquired using INS. Trajectory No. . Estimated errors (m) . Length (m) . Area enclosed (m2) . Min . Max . Average . SD . 1 0.01 0.93 0.33 0.29 869 20 660 2 0.01 0.62 0.34 0.18 1384 57 994 3 0.01 0.58 0.25 0.18 1246 52 763 Trajectory No. . Estimated errors (m) . Length (m) . Area enclosed (m2) . Min . Max . Average . SD . 1 0.01 0.93 0.33 0.29 869 20 660 2 0.01 0.62 0.34 0.18 1384 57 994 3 0.01 0.58 0.25 0.18 1246 52 763 Open in new tab Table 4 Basic statistical and geometrical characteristics of the reference trajectories acquired using INS. Trajectory No. . Estimated errors (m) . Length (m) . Area enclosed (m2) . Min . Max . Average . SD . 1 0.01 0.93 0.33 0.29 869 20 660 2 0.01 0.62 0.34 0.18 1384 57 994 3 0.01 0.58 0.25 0.18 1246 52 763 Trajectory No. . Estimated errors (m) . Length (m) . Area enclosed (m2) . Min . Max . Average . SD . 1 0.01 0.93 0.33 0.29 869 20 660 2 0.01 0.62 0.34 0.18 1384 57 994 3 0.01 0.58 0.25 0.18 1246 52 763 Open in new tab Experiment 2 Reference trajectories characteristics Measures describing the accuracy of reference trajectories based on single points’ errors estimated by the processing software, and geometrical characteristics of the trajectories (length, area enclosed) are reported in Table 4. Mean values of estimated errors were below 40 cm. The 95 per cent confidence interval (mean + 2 SDs) is under 1 m in all cases. Maxima are also less than 1 m. These errors of the reference trajectories must be considered when interpreting the results of the other tested devices. Trajectory accuracy The accuracy of trajectories was evaluated based on their shift from the reference trajectory (using buffers around the reference). Percentual shares belonging to individual buffers were calculated for trajectories T1–T3 (Supplementary Figures 4–6) and overall (Figure 9). The Trimble Nomad 5 is the most accurate in this task, with 84.4 per cent of the trajectories falling into the 1- and 2-m buffer. The average shift for this device was 1.23 m. Xiaomi Mi 8 with a 1.99-m average shift and 58 per cent of trajectories in the first two buffers is the second most accurate. The average shift of Lenovo Phab 2 Pro was 2.90 m with ~45 per cent of trajectories in the first two buffers, followed by LG G2 with 3.65 m and 38.7 per cent. The trajectories of Lenovo A5000 were shifted by 4.73 m on average. The worst results were achieved by the GNS2000 with an average shift of 5.98 m. Although this device was among the most accurate devices in the first experiment, its’ trajectories were highly deformed, with 17.8 per cent of the sections of the trajectories outside the 10-m buffer and a maximal shift of almost 38 m. Reliability of derived lengths and acreages The reliability of metrics derived from the trajectories – i.e. tracks’ lengths and enclosed acreages – was evaluated and their relative errors are reported in Table 5. Generally, the lengths and areas are overestimated, with the exception of the areas T1, T3 and few other values. Overall, the errors rarely exceed 10 per cent. This is the case for all lengths measured by GNS2000 and the length of the T1 trajectory measured by Lenovo Phab 2 Pro. For GNS2000, this can be related to the poor accuracy of the trajectory. In general, good trajectory accuracy expectedly leads to good results for lengths and areas. However, this is not true reciprocally. For example, the T2 area determined by GNS2000 has an error of 0.5 per cent, but the trajectory accuracy as well as the trajectory length show large errors. The same applies to the T3 length measured by Lenovo A5000, where the error is 0.2 per cent, but the area is underestimated by 9.4 per cent and the average shift of the trajectory is 5.3 m. Overall, lengths and areas cannot be considered good measures to evaluate trajectory data. Table 5 Areas and lengths determined by tested devices and their relative errors for individual trajectories (T1–T3). Device . Trajectory T1 . Trajectory T2 . Trajectory T3 . Area (m2) . Length (m) . Area (m2) . Length (m) . Area (m2) . Length (m) . Reference (INS) 20 660 869 57 994 1384 52 763 1246 Trimble Nomad 5 20 168 (−2.4%) 869 (0.0%) 58 162 (+0.3%) 1368 (−1.2%) 52 257 (−1.0%) 1210 (−2.9%) Xiaomi Mi 8 20 254 (−2.0%) 913 (+5.1%) 59 039 (+1.8%) 1467 (+6.0%) 51 458 (−2.5%) 1280 (+2.7%) GNS2000 20 473 (−0.9%) 957 (+10.1%) 58 304 (+0.5%) 1629 (+17.7%) 49 091 (−7.0%) 1458 (+17.0%) LG G2 20 732 (+0.3%) 928 (+6.8%) 60 514 (+4.3%) 1410 (+1.9%) 51 456 (−2.5%) 1288 (+3.4%) Lenovo A5000 20 000 (−3.2%) 898 (+3.3%) 54 226 (−6.5%) 1404 (+1.4%) 47 787 (−9.4%) 1248 (+0.2%) Lenovo Phab 2 Pro 22 595 (+9.4%) 982 (+13.0%) 58 490 (+0.9%) 1422 (+2.7%) 52 563 (−0.4%) 1317 (+5.7%) Device . Trajectory T1 . Trajectory T2 . Trajectory T3 . Area (m2) . Length (m) . Area (m2) . Length (m) . Area (m2) . Length (m) . Reference (INS) 20 660 869 57 994 1384 52 763 1246 Trimble Nomad 5 20 168 (−2.4%) 869 (0.0%) 58 162 (+0.3%) 1368 (−1.2%) 52 257 (−1.0%) 1210 (−2.9%) Xiaomi Mi 8 20 254 (−2.0%) 913 (+5.1%) 59 039 (+1.8%) 1467 (+6.0%) 51 458 (−2.5%) 1280 (+2.7%) GNS2000 20 473 (−0.9%) 957 (+10.1%) 58 304 (+0.5%) 1629 (+17.7%) 49 091 (−7.0%) 1458 (+17.0%) LG G2 20 732 (+0.3%) 928 (+6.8%) 60 514 (+4.3%) 1410 (+1.9%) 51 456 (−2.5%) 1288 (+3.4%) Lenovo A5000 20 000 (−3.2%) 898 (+3.3%) 54 226 (−6.5%) 1404 (+1.4%) 47 787 (−9.4%) 1248 (+0.2%) Lenovo Phab 2 Pro 22 595 (+9.4%) 982 (+13.0%) 58 490 (+0.9%) 1422 (+2.7%) 52 563 (−0.4%) 1317 (+5.7%) Bold indicates errors exceeding 10 per cent. Open in new tab Table 5 Areas and lengths determined by tested devices and their relative errors for individual trajectories (T1–T3). Device . Trajectory T1 . Trajectory T2 . Trajectory T3 . Area (m2) . Length (m) . Area (m2) . Length (m) . Area (m2) . Length (m) . Reference (INS) 20 660 869 57 994 1384 52 763 1246 Trimble Nomad 5 20 168 (−2.4%) 869 (0.0%) 58 162 (+0.3%) 1368 (−1.2%) 52 257 (−1.0%) 1210 (−2.9%) Xiaomi Mi 8 20 254 (−2.0%) 913 (+5.1%) 59 039 (+1.8%) 1467 (+6.0%) 51 458 (−2.5%) 1280 (+2.7%) GNS2000 20 473 (−0.9%) 957 (+10.1%) 58 304 (+0.5%) 1629 (+17.7%) 49 091 (−7.0%) 1458 (+17.0%) LG G2 20 732 (+0.3%) 928 (+6.8%) 60 514 (+4.3%) 1410 (+1.9%) 51 456 (−2.5%) 1288 (+3.4%) Lenovo A5000 20 000 (−3.2%) 898 (+3.3%) 54 226 (−6.5%) 1404 (+1.4%) 47 787 (−9.4%) 1248 (+0.2%) Lenovo Phab 2 Pro 22 595 (+9.4%) 982 (+13.0%) 58 490 (+0.9%) 1422 (+2.7%) 52 563 (−0.4%) 1317 (+5.7%) Device . Trajectory T1 . Trajectory T2 . Trajectory T3 . Area (m2) . Length (m) . Area (m2) . Length (m) . Area (m2) . Length (m) . Reference (INS) 20 660 869 57 994 1384 52 763 1246 Trimble Nomad 5 20 168 (−2.4%) 869 (0.0%) 58 162 (+0.3%) 1368 (−1.2%) 52 257 (−1.0%) 1210 (−2.9%) Xiaomi Mi 8 20 254 (−2.0%) 913 (+5.1%) 59 039 (+1.8%) 1467 (+6.0%) 51 458 (−2.5%) 1280 (+2.7%) GNS2000 20 473 (−0.9%) 957 (+10.1%) 58 304 (+0.5%) 1629 (+17.7%) 49 091 (−7.0%) 1458 (+17.0%) LG G2 20 732 (+0.3%) 928 (+6.8%) 60 514 (+4.3%) 1410 (+1.9%) 51 456 (−2.5%) 1288 (+3.4%) Lenovo A5000 20 000 (−3.2%) 898 (+3.3%) 54 226 (−6.5%) 1404 (+1.4%) 47 787 (−9.4%) 1248 (+0.2%) Lenovo Phab 2 Pro 22 595 (+9.4%) 982 (+13.0%) 58 490 (+0.9%) 1422 (+2.7%) 52 563 (−0.4%) 1317 (+5.7%) Bold indicates errors exceeding 10 per cent. Open in new tab Table 6 Horizontal and vertical errors (in meters) of average positions acquired using raw GNSS data according to processing modification and specific conditions. Processing settings . Direction . Conditions . Oopen area . Mature mixeda . Mature coniferousb . Post-disturbance openingc . Inner edged . L1 GPS + GLO horizontal 0.34 83.48 11.25 46.44 0.99 vertical 0.03 74.67 1.99 5.61 -4.63 L1L5 GPS + GLO horizontal 0.50 58.42 11.7 53.5 17.29 vertical 0.03 36.07 -2.82 10.97 30.01 L1 all horizontal 0.06 50.38 6.4 34.77 4.98 vertical -0.05 48.84 3.81 3.12 -6.86 L1L5 all Horizontal 0.07 30.53 3.48 33.72 16.54 Vertical -0.04 13.28 2.44 5.51 28.48 Instantaneous5 Horizontal 0.31 4.77 2.33 2.83 2.71 L1L5 all Vertical 0.53 14.22 20.46 23.38 6.73 Processing settings . Direction . Conditions . Oopen area . Mature mixeda . Mature coniferousb . Post-disturbance openingc . Inner edged . L1 GPS + GLO horizontal 0.34 83.48 11.25 46.44 0.99 vertical 0.03 74.67 1.99 5.61 -4.63 L1L5 GPS + GLO horizontal 0.50 58.42 11.7 53.5 17.29 vertical 0.03 36.07 -2.82 10.97 30.01 L1 all horizontal 0.06 50.38 6.4 34.77 4.98 vertical -0.05 48.84 3.81 3.12 -6.86 L1L5 all Horizontal 0.07 30.53 3.48 33.72 16.54 Vertical -0.04 13.28 2.44 5.51 28.48 Instantaneous5 Horizontal 0.31 4.77 2.33 2.83 2.71 L1L5 all Vertical 0.53 14.22 20.46 23.38 6.73 L1, L5 – used frequencies; GPS + GLO, all – used systems; a–dLabelling according to Figure 2. 5Ambiguity resolution set to ‘Instantaneous’ instead of ‘Fix and hold’ in all other modifications. Open in new tab Table 6 Horizontal and vertical errors (in meters) of average positions acquired using raw GNSS data according to processing modification and specific conditions. Processing settings . Direction . Conditions . Oopen area . Mature mixeda . Mature coniferousb . Post-disturbance openingc . Inner edged . L1 GPS + GLO horizontal 0.34 83.48 11.25 46.44 0.99 vertical 0.03 74.67 1.99 5.61 -4.63 L1L5 GPS + GLO horizontal 0.50 58.42 11.7 53.5 17.29 vertical 0.03 36.07 -2.82 10.97 30.01 L1 all horizontal 0.06 50.38 6.4 34.77 4.98 vertical -0.05 48.84 3.81 3.12 -6.86 L1L5 all Horizontal 0.07 30.53 3.48 33.72 16.54 Vertical -0.04 13.28 2.44 5.51 28.48 Instantaneous5 Horizontal 0.31 4.77 2.33 2.83 2.71 L1L5 all Vertical 0.53 14.22 20.46 23.38 6.73 Processing settings . Direction . Conditions . Oopen area . Mature mixeda . Mature coniferousb . Post-disturbance openingc . Inner edged . L1 GPS + GLO horizontal 0.34 83.48 11.25 46.44 0.99 vertical 0.03 74.67 1.99 5.61 -4.63 L1L5 GPS + GLO horizontal 0.50 58.42 11.7 53.5 17.29 vertical 0.03 36.07 -2.82 10.97 30.01 L1 all horizontal 0.06 50.38 6.4 34.77 4.98 vertical -0.05 48.84 3.81 3.12 -6.86 L1L5 all Horizontal 0.07 30.53 3.48 33.72 16.54 Vertical -0.04 13.28 2.44 5.51 28.48 Instantaneous5 Horizontal 0.31 4.77 2.33 2.83 2.71 L1L5 all Vertical 0.53 14.22 20.46 23.38 6.73 L1, L5 – used frequencies; GPS + GLO, all – used systems; a–dLabelling according to Figure 2. 5Ambiguity resolution set to ‘Instantaneous’ instead of ‘Fix and hold’ in all other modifications. Open in new tab Practical problems and sources of errors encountered in Experiments 1 and 2 During the field survey and processing of the GNSS data, we encountered some typical issues. Drift (Figure 10a) – position drift was evident especially during longer periods when the devices were stationary. This issue was evident during the first experiment, where it was impossible to record the point position on all devices at the same moment. The drift can cause both increase and decrease of the distance from the reference. However, according to our experience, the initial position was better and the drift degraded the solution in most cases. In practice, when a single device is used for positioning, the influence of this issue can be minimized. Delay (Figure 10b) – a delay in position update was experienced especially when the distances between points were short and signal reception was complicated. The receiver does not recognize any change of the position and a further point is recorded with the position of the previous point or very close to it. This problem is more significant when measuring single points rather than continuous trajectories. Outage (Figure 10c) – although the occurrence of GNSS signal outages was rare, we experienced some especially for older receivers with less systems (constellations) applied. The occurrence of the outage is easy to distinguish on the trajectories – it is characterized by long, straight sections of the trajectory with sharp edges or even by gaps. This problem can influence both the single point measurements and the trajectory recording and can be significant, when the GNSS receivers (especially the older ones) are used under dense forest canopy combined with other GNSS adverse conditions (e.g. narrow valleys, etc.). Figure 9 Open in new tabDownload slide Overall percentual shares of trajectories belonging to 1-m-wide buffers according to tested devices. This figure appears in colour in the online version of Forestry. Figure 9 Open in new tabDownload slide Overall percentual shares of trajectories belonging to 1-m-wide buffers according to tested devices. This figure appears in colour in the online version of Forestry. Figure 10 Open in new tabDownload slide Typical issues encountered during the GNSS measurements in forests: (a) position drift, (b) delay in position change and (c) outage of GNSS measurements. Green triangles represent reference. This figure appears in colour in the online version of Forestry. Figure 10 Open in new tabDownload slide Typical issues encountered during the GNSS measurements in forests: (a) position drift, (b) delay in position change and (c) outage of GNSS measurements. Green triangles represent reference. This figure appears in colour in the online version of Forestry. These problems – especially drift and delay – affected mainly the static measurement of individual points. During the trajectory measurements, the movement was fluent (without pauses), and if some delay occurred, it was unobservable due to the character of the measurement. This is probably one of the reasons why the accuracy of the dynamic measurements was better. Experiment 3 Raw measurements post-processing accuracy The horizontal and vertical errors calculated for the positions derived from the raw GNSS data collected with the Xiaomi Mi 8 are reported in Table 6. The difference between accuracies achieved under open-area conditions and, in the forest, is obvious. In the forest, hardly any trend is visible when comparing the differing processing options and stand characteristics. When using the standard ‘Fix and hold’ ambiguity resolution strategy, the errors range from meters to tens of meters. There is just a single measurement with sub-meter error (0.99 m); however, the maximum is 83.48 m. Setting the ambiguity resolution to ‘Instantaneous’ significantly reduces the variability of the individual positions, and the horizontal errors of the average positions are in the range of meters. However, this improvement is visible only under sub-optimal conditions in the forest (errors between 2.33 and 4.77 m). Comparison of results achieved using ‘Fix and hold’ and ‘Instantaneous’ strategy under open-area conditions shows that under these condition, ‘Fix and hold’ provides much better results (0.07 vs 0.31 m using all systems and both frequencies). A clear improvement caused by the addition of newer satellite systems (Galileo, Beidou, QZSS) to the traditional ones (GPS, GLONASS) is evident for the open area and most of the forest settings. In the open area, horizontal errors decrease from decimetres to centimetres, vertical errors are in the range of centimetres for both constellations. In the forest, except for the inner edge of a mature mixed stand, horizontal errors decrease but are still in ranges of tens of meters with exception of the mature coniferous stand, where the accuracy was the highest. Regarding the addition of the second frequency, no clear benefit was evident; in fact, the solutions using both frequencies show worse horizontal errors under the open-area condition. The best horizontal accuracy in the open area was achieved using single-frequency data from all available systems. Under forest conditions, there are cases where the second frequency brought higher accuracy (e.g. mature mixed stand), but also the opposite ones (inner edge of a mature mixed stand). Overall, the horizontal error of 0.06 m with a vertical error also in the range of centimetres suggests a great potential of raw GNSS data measured using smartphones under open-area condition. Errors in the forest are very high and variable, hardly reaching even the results of our real-time tests. Graphical outputs together with more detailed statistical measures related to individual solutions and modifications are available as Supplementary Figure 7. Carrier-to-noise density C/N0 and signals observations quantity The basic statistical measures of the carrier-to-noise density for the open-area point and the point in the mature mixed forest are shown in Figure 11 and in Supplementary Figures 8–10 for other points used for the raw GNSS data measurements. The figures also contain numbers of observations for particular GNSS signals. Figure 11 Open in new tabDownload slide Minima, maxima, SDs and means of carrier-to-noise density (C/N0) for measurements under open area and mature mixed forest conditions, and numbers of observations according to GNSS signals. This figure appears in colour in the online version of Forestry. Figure 11 Open in new tabDownload slide Minima, maxima, SDs and means of carrier-to-noise density (C/N0) for measurements under open area and mature mixed forest conditions, and numbers of observations according to GNSS signals. This figure appears in colour in the online version of Forestry. Under open-area conditions, the mean carrier-to-noise density is 34.27 dB Hz. With decreasing open sky availability, mean C/N0 values decrease to 29.35 (post-disturbance opening) and 27.15 dB Hz (inner edge of forest stand). Under condition with nearly full canopy, the values are similar for the mixed and coniferous forest stand, i.e. 26.56 and 26.39 dB Hz. In these cases, the signal strength loss exceeds 20 per cent of the open-area value. The difference between L1(E1) and L5(E5) C/N0 values is 6.32 dB Hz for the GPS system and 5.10 dB Hz for Galileo considering open-area conditions. Although the significance of these differences was not statistically confirmed (ANOVA), such decrease in the signal strength could have practical implications. In the forest, the C/N0 differences between L1(E1) and L5(E5) signals exceeded 2.5 dB Hz only for Galileo under the conditions of inner edge of forest stand. The overall number of GNSS signal observations recorded under the open-area conditions, taking the 1-h observation period and 10-degrees elevation mask into the account, was 110 262. Subsequently, it decreases to 86 940 for the inner edge of forest stand and ~76 000 for the post-disturbance opening and mature forest stands (coniferous and mixed). This decrease was observed despite the increase of theoretically available satellites related to the movement of satellites. The ratio between GPS L5 and L1 signal observations varied from ~1:4 to 1:3 under all tested conditions. For Galileo signals, the number of E5 observations was higher than the number of E1 observations in all cases. Generally, the forest conditions caused a 20 per cent decrease in the signal strength and a 30 per cent decrease in observations available for position processing. Discussion Results of any GNSS study examining the accuracy achievable in forests are very hard to generalize, especially when speaking about smartphones. Just during the second quarter of 2019, worldwide smartphone vendors shipped a total of 332.2 million (IDC, 2019a) smartphones, resulting in tens of types and hardware configurations of the devices. Current smartphone GNSS receivers are characterized by supporting multi-constellation approaches and in some cases also dual-frequency measurements. However, newer hardware does not necessarily mean better accuracy, as was reported by Szot et al. (2019). Besides the hardware, also the version of the operating system and the application used for GNSS measurements can influence the accuracy (Bauer, 2013; Tomaštík et al., 2017). Although the numerical values cannot be generalized, the Xiaomi Mi 8 with the dual-frequency receiver achieved results better than other smartphones in this study. In terms of statistical significance, it was comparable to the single-frequency mapping-grade receiver. However, besides the clearly higher number of available signals, the contribution of the second frequency is hard to measure. It was impossible to quantify it in our first and second experiments, because it was impossible to process the first and the second frequency individually. The results obtained by post-processing the raw GNSS data, as examined in the third experiment, do not show any significant contribution of the second frequency. This can be partially explained by the lower availability of the second frequency, because just approximately one-third of currently operational GPS satellites (Block IIF and GPS III/IIIF) are able to broadcast the L5 frequency (National Coordination Office for Space-Based Positioning, Navigation and Timing, 2019). Dabove and Di Pietra (2019) reported a horizontal and vertical precision of 2 cm for the combined L1 + L5 frequency-based GNSS measurements acquired using Xiaomi Mi8, while they found a precision of 3-cm horizontally and 2-cm vertically for the L1 frequency alone. It is necessary to mention that with exception of the Huawei P20 lite, other tested smartphones were one to three generations older than Xiaomi Mi 8. Thus, the clear increase in accuracy of the new generation smartphones may be credited firstly to the higher number of available systems and only subsequently to the second frequency. In our test, the number of available satellite signals for the Xiaomi Mi 8 (GPS, GLONASS, Galileo, Beidou, QZSS, L1 + L5 signals) was more than three times higher than for the Lenovo A5000 (GPS L1 signal only). We did not use operating system other than Android, as this OS is dominant with a market share exceeding 85 per cent (IDC, 2019b), and smartphones with other operating systems do not provide an option to record raw GNSS data. When comparing the current results of the first experiment with the results of our previous study using a similar approach (Tomaštík et al., 2017), a clear increase in accuracy is observed only for the newest multi-constellation, dual- frequency receiver. This can be partially caused by the fact that most of the tested smartphones were mid-range devices (in terms of both price and hardware specifications), which on the other hand are the most frequently used. The accuracies observed in the first experiment suggest that smartphones may be used for the measurement of various point features (e.g. important landmarks, test plots centres, etc.) as well as for navigation purposes under forest conditions. However, even with the newest tested device, the accuracy in a forest during the leaf-on season was not better than 5 m. This is important, especially in cases where strict accuracy criteria are applied in the field of forestry mapping. On the other hand, mapping-grade receivers that do not use differential corrections acquire very similar levels of accuracy and could be replaced by mobile ‘smart’ devices (Lee et al., 2020). With regard to the previous study (Tomaštík et al., 2017), the newer generation of smartphones provides better accuracy under forest conditions. However, under open-area conditions, the accuracy of real-time, unprocessed measurements probably already reached its limits with an accuracy of ~2 m. A forest clearly degrades the quality of received GNSS signals. This is obvious in our first experiment, where the errors in the forest are several times higher than in the open area, and especially in the third experiment, where transition into the forest caused an increase in horizontal errors from decimetres to meters and even tens of meters, thus limiting many practical applications. In the first experiment, we tried to evaluate the relation between differing forest stand settings and horizontal accuracy. Only two significant correlations between the openness of the forest and the accuracy were found. For Trimble Nomad 5, the accuracy increased with increasing openness (R = −0.498) under leaf-off conditions. In a direct contradiction with this, the accuracy of Xiaomi Mi 8 decreased with increasing openness under leaf-on conditions (R = 0.549). We consider this result illogical and recommend to repeat such an analysis with more sample points. Earlier studies attempted to explain GNSS accuracies with typical forest stand metrics such as basal area, DBH or tree height but failed to find high correlations. For example, the best model designed by Frank and Wing (2014) explained only 14 per cent of the variation in the positional accuracy, taking the basal area, terrain slope, horizontal dilution of precision, and receiver settings into account. Naesset (2001) reported R2 values of 0.227 for basal area and 0.190 for tree height using a survey-grade receiver. In another study (Zimbelman and Keefe, 2018), RMSE increased with tree height but varied indirectly with quadratic mean diameter. Wing et al. (2005) reported accuracies of 5 m within open sky settings, 7 m in a young forest and 10 m in a closed canopy setting for consumer-grade receivers. Bettinger and Merry (2012) took total tree count and basal area within a 5-m radius plot around a test point and found weak correlation (Pearson’s r from −0.30 to 0.30) with the positional error. Aside from examining the positional accuracy of point measurements, we tested also the applicability of smartphone GNSS positioning for recording trajectories and subsequently derived geometric measures (lengths, areas). Ucar et al. (2014) conducted a similar experiment with a 0.90-ha test plot in an oak-hickory forest and reported an average area of agreement of ~93 per cent during the leaf-off season and 84 per cent during the leaf-on season using a recreational-grade GNSS receiver. Although our measurements were conducted during the leaf-on season, the differences of areas did not exceed ±10 per cent. For most examined devices, the mean shift of trajectories (second experiment) was lower than the mean positional error achieved on points (first experiment). Although these results are not directly comparable, we conclude that smartphone receivers are better suited for dynamic applications. On the other hand, the GNS2000 receiver, which provided good static results, was not able to provide acceptable trajectories. In this experiment, we used the reference obtained using a pedestrian INS. The main advantage of this approach is that we had the reference in every moment of the measurements because the trajectories were collected simultaneously. On the other hand, the INS accuracy under forest canopies reaches only a sub-meter level, which was reported also in other studies (Soloviev et al., 2012; Kaartinen et al., 2015). This must be taken into account when interpreting the results. The second experiment is relevant for several forestry-related tasks such as personal and machinery tracking, but particularly also to delineate and map logged or disturbed areas as well as other parts of forests, where the application allows a somewhat lower degree of accuracy of ~2–5 m. We consider our study one of the first attempting to make use of raw GNSS data provided by smartphone receivers in forests. Multiple studies have already proven the ability to achieve sub-meter or even sub-decimetre accuracy using these measurements under open-area conditions (Lu et al., 2018; Banville et al., 2019; Dabove and Di Pietra, 2019; Robustelli et al., 2019; Wu et al., 2019). Our results also show high accuracies under such conditions, reaching centimetre-level accuracy in the best cases. However, our results under forest conditions are very variable and, in most cases, worse than real-time measurements without post-processing. The only possibility to achieve stable accuracies under 5 m was to set the ‘Integer ambiguity resolution’ to ‘Instantaneous’. This option deactivates the Kalman filter to predict changes between epochs, which improved the accuracies in the forest, but degraded the solutions under open-area conditions and should according to our results hence not be used for non-forested areas. At this point, we do not have a clear explanation for this. However, we acknowledge that our dataset is limited and this issue should be subject to future investigations. Generally, the challenges complicating further improvements in the smartphone positioning accuracy can be divided into hardware- and software-related. From the hardware point of view, the main problem is the applied GNSS antenna. The need of miniaturization leads to the application of primitive, linearly polarized patch antennas (Lu et al., 2018; Robustelli et al., 2019). To be able to provide positions in every pose of the smartphone, the antennas are omnidirectional (Banville et al., 2019), thus increasing the proneness to interference and multipath effect even more (Paziewski et al., 2019). To partially address this problem, we used a ground plane to mitigate the effects of signals reflected from the ground. A decrease in signal strength in range of ~10 dB Hz was reported for comparison of smartphone antennas and survey-grade hardware (Zhang et al., 2018; Liu et al., 2019). Moreover, our results suggest that the signal is ~20 per cent weaker in a forest compared with open-area conditions. With that regard, authors suggest preferring signal-strength-based weighting over the standardly used elevation masking (Banville et al., 2019; Paziewski et al., 2019) and the use of C/N0 thresholds to acquire valid measurements (Massarweh et al., 2019). Studies evaluating performance of smartphone GNSS receivers with an external higher grade antenna have been conducted (Siddakatte et al., 2017; Lachapelle et al., 2018; Geng and Li, 2019), but the experimental design, i.e. complicated connection between the antenna and receiver, prohibits the use by an average user. We are not aware of any practically applicable solution that combines the internal receiver of a smartphone with an external antenna. Apart from hardware problems, also software must be adapted to process data from low-cost receivers. Android 9 allows the user to switch off the so-called duty cycling (‘Developer options: Force full GNSS measurements’), which was meant to provide better energy effectiveness, but caused discontinuities in the reception of raw GNSS data (Lu et al., 2018). The Demo5 version of the RTKLib software, used in this study, is actively developed with emphasis on low-cost receivers and their specifics. Besides offline software, also some automated online processing services are being adapted to the processing of low-cost GNSS receiver data – e.g. Canadian CSRS-PPP (Banville et al., 2019). Although the possibilities of using raw GNSS data from smartphone receivers are open, these are often limited by manufacturer’s decisions. For example, the newer generation of the Xiaomi ‘flagship’ smartphone, the Xiaomi Mi 9, cannot provide carrier-phase data, thus restricting some post-processing procedures. If a user wants to use raw data, the possibility of providing the necessary types of the data must be checked for every smartphone model individually. Future research in the field of raw GNSS measurements should examine how shortening of the observation period affects the positional accuracies of this latest generation of smartphones and how different forest conditions influence the positional accuracy under forest canopies. Conclusion Although forests represent a typical example of GNSS adverse conditions, GNSS measurements are frequently applied for forestry-related tasks. Low-cost receivers and especially smartphones are commonly used in this context and there is hence a need to understand whether new developments in smartphone GNSS technology improve the positional accuracy in forested areas. We examined the positional accuracy of a dual-frequency GNSS device compared with single-frequency GNSS devices for measuring point locations and for mapping trajectories. In our real-time experiments, the Xiaomi Mi 8 smartphone with the Broadcom BCM47755 dual-frequency receiver achieved accuracies close to a mapping-grade receiver and better than the tested single-frequency smartphone receivers. This accuracy improvement must be credited firstly to the availability of more GNSS systems (constellations) and secondly to the second frequency (L5) available through the new receiver. Generally, the higher number of available signals results not only in the better accuracy but also in better availability and robustness of positioning solutions. The RMSE for the Xiaomi Mi 8 decreased from 6.13 m during the leaf-on season to 4.10 m for the leaf-off season, and finally to 2.23 m under open-area conditions. Test of trajectories tracking showed promising results. The error of the determined areas and lengths for all tested devices rarely exceeded 10 per cent and showed good spatial agreement with the reference delineation using an INS. Furthermore, we examined whether the recently added option to collect raw GNSS data by a smartphone can further improve the measurements of point locations. Our test based on the post-processing of raw GNSS data provided by the smartphone receiver showed substantial differences between the accuracies obtainable in forests and under conditions with optimal GNSS signal reception. Although the best open-area results achieved centimetre-level accuracy, thus allowing even geodetic applications, the accuracies in the forest varied from meters to tens of meters, which restricts many practical applications. To address this problem, changes in hardware (especially better antennas) and software (better adaptation to lower quality data provided by low-cost receivers) must be made. The known accuracy ranges could enable combinations of smartphone positioning with other tasks performed by smartphones (from simple evidence applications to emerging remote sensing methods), which can be highly beneficial for forestry research and practice. Data availability statement Partial data (raw GNSS measurements) are available in Mendeley Data service, at https://dx.doi.org/10.17632/knwttdwv5b.1. Other data can be shared on a reasonable request to the corresponding author. Acknowledgements Authors would like to thank Rút Tomaštíková for help during the field survey, Tim Everett for active software support and Zuzana Danihelová for language suggestions. We also would like to thank the editors (especially Fabian Fassnacht) and reviewers for their effort and valuable comments. Funding Scientific Grant Agency (VEGA) of the Ministry of Education, Science, Research and Sport of the Slovak Republic and the Slovak Academy of Sciences (grant number 1/0868/18) (Innovative techniques for mapping of anthropogenic and natural landforms applicable in survey of landscape status) and (grant number 1/0335/20) (A multi-camera system prototype as a tool for creating a highly detailed model of individual trees and forest stands). Conflict of interest statement None declared. References Apostol , B. , Chivulescu , S., Ciceu , A., Petrila , M., Pascu , I.S., Apostol , E.N. et al. 2018 Data collection methods for forest inventory: a comparison between an integrated conventional equipament and terrestrial laser scanning . Ann. For. Res. 61 , 189 – 202 . doi: 10.15287/afr.2018.1189 . Google Scholar Crossref Search ADS WorldCat Banville , S. , Lachapelle , G., Ghoddousi-Fard , R., and Gratton , P. 2019 Automated processing of low-cost GNSS receiver data. In Institute of Navigation GNSS+ 2019 Conference, Miami (Miami) . doi: 10.33012/2019.16972 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Crossref Bauer , C. 2013 On the (in-)accuracy of GPS measures of smartphones: a study of running tracking applications. In 11th International Conference on Advances in Mobile Computing & Multimedia (MoMM2013). 2–4 December 2013, Vienna, Austria , pp. 335 – 340 . Bettinger , P. and Merry , K.L. 2012 Influence of the juxtaposition of trees on consumer-grade GPS position quality . Math. Comput. For. Nat. Sci. 4 , 81 – 91 . Google Scholar OpenURL Placeholder Text WorldCat Bettinger , P. , Merry , K., Bayat , M. and Tomaštík , J. 2019 GNSS use in forestry—a multi-national survey from Iran, Slovakia and southern USA . Comput. Electron. Agric. 158 , 369 – 383 . doi: 10.1016/j.compag.2019.02.015 . Google Scholar Crossref Search ADS WorldCat Bianchi , S. , Cahalan , C., Hale , S. and Gibbons , J.M. 2017 Rapid assessment of forest canopy and light regime using smartphone hemispherical photography . Ecol. Evol. 7 , 10556 – 10566 . doi: 10.1002/ece3.3567 . Google Scholar Crossref Search ADS PubMed WorldCat Brach , M. , Stereńczak , K., Bolibok , L., Kwaśny , Ł., Krok , G. and Laszkowski , M. 2019 Impacts of forest spatial structure on variation of the multipath phenomenon of navigation satellite signals . Folia For. Pol. 61 , 3 – 21 . doi: 10.2478/ffp-2019-0001 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Dabove , P. and Di Pietra , V. 2019 Single-baseline RTK positioning using dual-frequency GNSS receivers inside smartphones . Sensors 19 , 4302 . doi: 10.3390/s19194302 . Google Scholar Crossref Search ADS WorldCat Dash , J. , Pont , D., Watt , M.S., Dash , J., Pont , D., Brownlie , R. et al. 2016 Remote sensing for precision forestry . NZ J. For. 15 – 24 . Google Scholar OpenURL Placeholder Text WorldCat van Diggelen , F. 2009 A-GPS: Assisted GPS, GNSS, and SBAS . Artech House , Boston , pp. 350 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Everett , T. 2019 rtklibexplorer—Exploring Precision GPS/GNSS with RTKLIB Open Source Software and Low-cost u-blox GNSS Receivers . https://rtklibexplorer.wordpress.com/ Fassnacht , F.E. , Mangold , D., Schäfer , J., Immitzer , M., Kattenborn , T., Koch , B. et al. 2017 Estimating stand density, biomass and tree species from very high resolution stereo-imagery-towards an all-in-one sensor for forestry applications? Forestry 90 , 613 – 631 . doi: 10.1093/forestry/cpx014 . Google Scholar Crossref Search ADS WorldCat Fortunato , M. , Ravanelli , M. and Mazzoni , A. 2019 Real-time geophysical applications with android GNSS raw measurements . Remote Sens. 11 , 2113 . doi: 10.3390/rs11182113 . Google Scholar Crossref Search ADS WorldCat Frank , J. and Wing , M.G. 2014 Balancing horizontal accuracy and data collection efficiency with mapping-grade GPS receivers . Forestry 87 , 389 – 397 . doi: 10.1093/forestry/cpt054 . Google Scholar Crossref Search ADS WorldCat Geng , J. and Li , G. 2019 On the feasibility of resolving android GNSS carrier-phase ambiguities . J. Geod. doi: 10.1007/s00190-019-01323-0 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Google 2019 Raw GNSS Measurements|Android Developers . https://developer.android.com/guide/topics/sensors/gnss GSA Raw Measurements Task Force 2017 Using Gnss Raw Measurements On Android Devices—Towards Better Location Performance in Mass Market Applications . European GNSS Agency doi: 10.2878/449581 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Crossref Guerra-Hernández , J. , González-Ferreiro , E., Monleón , V.J., Faias , S.P., Tomé , M. and Díaz-Varela , R.A. 2017 Use of multi-temporal UAV-derived imagery for estimating individual tree growth in Pinus pinea stands . Forests 8 . doi: 10.3390/f8080300 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Holopainen , M. , Vastaranta , M. and Hyyppä , J. 2014 Outlook for the next generation’s precision forestry in Finland . Forests 5 , 1682 – 1694 . doi: 10.3390/f5071682 . Google Scholar Crossref Search ADS WorldCat IDC 2019a IDC—Smartphone Market Share—Vendor . https://www.idc.com/promo/smartphone-market-share/vendor IDC 2019b IDC—Smartphone Market Share—OS . https://www.idc.com/promo/smartphone-market-share/os Irsigler , M. , Hein , G.W. and Eissfeiler , B. 2004 Multipath performance analysis for future GNSS signals . Proc. Natl. Tech. Meet. Inst. Navig. 2004 , 225 – 238 . Google Scholar OpenURL Placeholder Text WorldCat Joseph , A. 2010 Measuring GNSS signal strength . Inside GNSS 20 , 20 – 25 . Google Scholar OpenURL Placeholder Text WorldCat Kaartinen , H. , Hyyppä , J., Vastaranta , M., Kukko , A., Jaakkola , A., Yu , X. et al. 2015 Accuracy of kinematic positioning using global satellite navigation systems under forest canopies . Forests 6 , 3218 – 3236 . doi: 10.3390/f6093218 . Google Scholar Crossref Search ADS WorldCat Keefe , R.F. , Wempe , A.M., Becker , R.M., Zimbelman , E.G., Nagler , E.S., Gilbert , S.L. et al. 2019 Positioning methods and the use of location and activity data in forests . Forests 10 . doi: 10.3390/f10050458 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Kennedy , R. , McLeman , R., Sawada , M. and Smigielski , J. 2014 Use of smartphone technology for small-scale silviculture: a test of low-cost technology in eastern ontario . Small-scale For. 13 , 101 – 115 . doi: 10.1007/s11842-013-9243-5 . Google Scholar Crossref Search ADS WorldCat Lachapelle , G. , Gratton , P., Horrelt , J., Lemieux , E. and Broumandan , A. 2018 Evaluation of a low cost hand held unit with GNSS raw data capability and comparison with an android smartphone . Sensors (Switzerland) 18 . doi: 10.3390/s18124185 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Lee , T. , Bettinger , P., Cieszewski , C.J. and Gutierrez Garzon , A.R. 2020 The applicability of recreation-grade GNSS receiver (GPS watch, Suunto ambit peak 3) in a forested and an open area compared to a mapping-grade receiver (Trimble Juno T41) . PLoS One 15 , e0231532 . doi: 10.1371/journal.pone.0231532 . Google Scholar Crossref Search ADS PubMed WorldCat Liu , W. , Shi , X., Zhu , F., Tao , X. and Wang , F. 2019 Quality analysis of multi-GNSS raw observations and a velocity-aided positioning approach based on smartphones . Adv. Sp. Res. 63 , 2358 – 2377 . doi: 10.1016/j.asr.2019.01.004 . Google Scholar Crossref Search ADS WorldCat Lu , Y. , Ji , S., Chen , W., and Wang , Z. 2018 Assessing the performance of raw measurement from different types of smartphones. In 31st International Technical Meeting of the Satellite Division of the Institute of Navigation, ION GNSS+ 2018 . Institute of Navigation, Miami, Florida, pp. 304 – 322 . doi: 10.33012/2018.15881 Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Crossref Marzulli , M.I. , Raumonen , P., Greco , R., Persia , M. and Tartarino , P. 2020 Original article estimating tree stem diameters and volume from smartphone photogrammetric point clouds . For. An Int. J. For. Res. 93 , 411 – 429 . doi: 10.1093/forestry/cpz067 . Google Scholar Crossref Search ADS WorldCat Massarweh , L. , Darugna , F., Psychas , D., and Bruno , J. 2019 Statistical investigation of android GNSS data: case study using Xiaomi Mi 8 dual-frequency raw measurements. In 32nd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2019) . Institute of Navigation, Miami, Florida, pp. 3847 – 3861 . doi: 10.33012/2019.17072 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Crossref McGaughey , R.J. , Ahmed , K., Andersen , H.E. and Reutebuch , S.E. 2017 Effect of occupation time on the horizontal accuracy of a mapping-grade GNSS receiver under dense forest canopy . Photogramm. Eng. Remote Sensing 83 , 861 – 868 . doi: 10.14358/PERS.83.12.861 . Google Scholar Crossref Search ADS WorldCat Merry , K. and Bettinger , P. 2019 Smartphone GPS accuracy study in an urban environment . PLoS One 14 , e0219890 . doi: 10.1371/journal.pone.0219890 . Google Scholar Crossref Search ADS PubMed WorldCat Moudrý , V. , Gdulová , K., Fogl , M., Klápště , P., Urban , R., Komárek , J. et al. 2019 Comparison of leaf-off and leaf-on combined UAV imagery and airborne LiDAR for assessment of a post-mining site terrain and vegetation structure: Prospects for monitoring hazards and restoration success . Appl. Geogr. 104 , 32 – 41 . doi: 10.1016/j.apgeog.2019.02.002 . Google Scholar Crossref Search ADS WorldCat Murgaš , V. , Sačkov , I., Sedliak , M., Tunák , D. and Chudý , F. 2018 Assessing horizontal accuracy of inventory plots in forests with different mix of tree species composition and development stage . J. For. Sci. 64 , 478 – 485 . doi: 10.17221/92/2018-JFS . Google Scholar Crossref Search ADS WorldCat Naesset , E. 2001 Effects of differential single- and dual-frequency GPS and GLONASS observations on point accuracy under Forest canopies . Photogramm. Eng. Remote Sensing 69 , 1021 – 1026 . Google Scholar OpenURL Placeholder Text WorldCat NASA 2019 Daily Rinex files . ftp://cddis.gsfc.nasa.gov/gnss/data/campaign/mgex/daily/rinex3/2019/. National Coordination Office for Space-Based Positioning, Navigation, and T . 2019 GPS Space Segment . https://www.gps.gov/systems/gps/space/. Ordóñez Galán , C. , Rodríguez-Pérez , J.R., Martínez Torres , J. and García Nieto , P.J. 2011 Analysis of the influence of forest environments on the accuracy of GPS measurements by using genetic algorithms . Math. Comput. Model. 54 , 1829 – 1834 . doi: 10.1016/j.mcm.2010.11.077 . Google Scholar Crossref Search ADS WorldCat Paziewski , J. , Sieradzki , R. and Baryla , R. 2019 Signal characterization and assessment of code GNSS positioning with low-power consumption smartphones . GPS Solut. 23 , 1 – 12 . doi: 10.1007/s10291-019-0892-5 . Google Scholar Crossref Search ADS WorldCat Pérez , J.R. 2006 Comparison of GPS receiver accuracy and precision in forest environments . In Proceeding of XXIII FIG Congress , International Federation of Surveyors, Munich, Germany. Google Scholar OpenURL Placeholder Text WorldCat Piedallu , C. and Gégout , J.C. 2005 Effects of forest environment and survey protocol on GPS accuracy . Photogramm. Eng. Remote Sensing 71 , 1071 – 1078 . doi: 10.14358/PERS.71.9.1071 . Google Scholar Crossref Search ADS WorldCat QGIS Development Team 2019 QGIS Geographic Information System . http://qgis.osgeo.org/. Robustelli , U. , Baiocchi , V. and Pugliano , G. 2019 Assessment of dual frequency GNSS observations from a Xiaomi Mi 8 android smartphone and positioning performance analysis . Electron. 8 . doi: 10.3390/electronics8010091 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Romero-Andrade , R. , Zamora-Maciel , A., Uriarte-Adrián , J.d.J., Pivot , F. and Trejo-Soto , M.E. 2019 Comparative analysis of precise point positioning processing technique with GPS low-cost in different technologies with academic software . Meas. J. Int. Meas. Confed. 136 , 337 – 344 . doi: 10.1016/j.measurement.2018.12.100 . Google Scholar Crossref Search ADS WorldCat Sačkov , I. , Kulla , L. and Bucha , T. 2019 A comparison of two tree detection methods for estimation of forest stand and ecological variables from airborne LiDAR data in central european forests . Remote Sens. 11 . doi: 10.3390/rs11121431 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Schaefer , M. and Woodyer , T. 2015 Assessing absolute and relative accuracy of recreation-grade and mobile phone GNSS devices: a method for informing device choice . Area 47 , 185 – 196 . doi: 10.1111/area.12172 . Google Scholar Crossref Search ADS WorldCat Schwieger , V. 2009 High-sensitivity GPS—an availability, reliability and accuracy test . Bull. des Sci. Geogr. 23 , 12 . Google Scholar OpenURL Placeholder Text WorldCat Siddakatte , R. , Broumandan , A. and Lachapelle , G. 2017 Performance evaluation of smartphone gnss measurements with different antenna configurations. In Royal Institute of Navigation International Navigation Conference . Brighton, pp. 27 – 30 . https://schulich.ucalgary.ca/labs/position-location-and-navigation/files/position-location-and-navigation/siddakatte2017conference_c.pdf SKPOS 2019 Resort Transformation Service . https://zbgis.skgeodesy.sk/rts/sk/Transform. Šmelko , Š. 2007 Dendrometry . Technical University in Zvolen , Zvolen , pp. 401 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Soloviev , A. , Toth , C. and Grejner-Brzezinska , D. 2012 Performance of deeply integrated GPS/INS in dense forestry areas . J. Appl. Geod. 6 , 3 – 13 . doi: 10.1515/jag-2011-0005 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Szot , T. , Specht , C., Specht , M. and Dabrowski , P.S. 2019 Comparative analysis of positioning accuracy of Samsung Galaxy smartphones in stationary measurements . PLoS One 14 , e0215562 . doi: 10.1371/journal.pone.0215562 . Google Scholar Crossref Search ADS PubMed WorldCat Takasu , T. 2013 RTKLIB ver. 2.4.2 Manual . Tokyo University of Marine Science and Technology , Tokyo, Japan . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Taylor , S.E. , McDonald , T.P., Veal , M.W., Corley , F.W. and Grift , T.E. 2002 Precision forestry: operational tactics for today and tomorrow. In 25th Annual Meeting of the Council of Forest Engineers . Auburn University, Auburn, Alabama Vol. 6 . Tomaštík , J. 2019 Raw GNSS Data From Forest (Four Points) And Open Area (One Point) . Mendeley Data service doi: 10.17632/knwttdwv5b.1 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Crossref Tomaštík , J. , Tomaštík , J., Saloň , Š. and Piroh , R. 2017 Horizontal accuracy and applicability of smartphone GNSS positioning in forests . Forestry 90 , 187 – 198 . doi: 10.1093/forestry/cpw031 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Tuček , J. , Kardoš , M. and Tomaštík , J. 2016 First experience with pedestrian inertial navigation system application under forest conditions . Zpravy Lesn. Vyzk. 61 , 203 – 212 . Google Scholar OpenURL Placeholder Text WorldCat Ucar , Z. , Bettinger , P., Weaver , S., Merry , K.L. and Faw , K. 2014 Dynamic accuracy of recreation-grade GPS receivers in oak-hickory forests . Forestry 87 , 504 – 511 . doi: 10.1093/forestry/cpu019 . Google Scholar Crossref Search ADS WorldCat Vastaranta , M. , Latorre , E.G., Luoma , V., Saarinen , N., Holopainen , M. and Hyyppä , J. 2015 Evaluation of a smartphone app for forest sample plot measurements . Forests 6 , 1179 – 1194 . doi: 10.3390/f6041179 . Google Scholar Crossref Search ADS WorldCat White , J.C. , Coops , N.C., Wulder , M.A., Vastaranta , M., Hilker , T. and Tompalski , P. 2016 Remote sensing technologies for enhancing forest inventories: a review . Can. J. Remote Sens. 42 , 619 – 641 . doi: 10.1080/07038992.2016.1207484 . Google Scholar Crossref Search ADS WorldCat Wing , M.G. , Eklund , A. and Kellogg , L.D. 2005 Consumer-grade global positioning system (GPS) accuracy and reliability . J. For. 103 , 169 – 173 . doi: 10.1093/jof/103.4.169 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Wu , Q. , Sun , M., Zhou , C. and Zhang , P. 2019 Precise point positioning using dual-frequency GNSS observations on smartphone . Sensors (Switzerland) 19 . doi: 10.3390/s19092189 . Google Scholar OpenURL Placeholder Text WorldCat Crossref Zandbergen , P.A. and Barbeau , S.J. 2011 Positional accuracy of assisted GPS data from high-sensitivity GPS-enabled mobile phones . J. Navig. 64 , 381 – 399 . doi: 10.1017/S0373463311000051 . Google Scholar Crossref Search ADS WorldCat Zhang , K. , Jiao , F., and Li , J. 2018 The Assessment of GNSS Measurements from Android Smartphones. In China Satellite Navigation Conference (CSNC) 2018 Proceedings . J. Sun, C. Yang and S. Guo (eds). Springer , Singapore , pp. 147 – 157 . doi: 10.1007/978-981-13-0029-5_14 Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Zimbelman , E.G. and Keefe , R.F. 2018 Real-time positioning in logging: effects of forest stand characteristics, topography, and line-of-sight obstructions on GNSS-RF transponder accuracy and radio signal propagation . PLoS One 13 , 1 – 17 . doi: 10.1371/journal.pone.0191017. Google Scholar Crossref Search ADS WorldCat © The Author(s) 2020. Published by Oxford University Press on behalf of Institute of Chartered Foresters. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Advances in smartphone positioning in forests: dual-frequency receivers and raw GNSS data
JF - Forestry
DO - 10.1093/forestry/cpaa032
DA - 2021-03-04
UR - https://www.deepdyve.com/lp/oxford-university-press/advances-in-smartphone-positioning-in-forests-dual-frequency-receivers-Uc0RuGgbxd
SP - 292
EP - 310
VL - 94
IS - 2
DP - DeepDyve
ER -