How Accurate Is Cell Phone Gps Location

• Loading metrics

Open Access

Peer-reviewed

Research Article

Smartphone GPS accuracy report in an urban environment

• Krista Merry,
• Pete Bettinger

x

• Published: July 18, 2019
• https://doi.org/10.1371/journal.pone.0219890

Figures

Abstruse

An iPhone 6 using the Avenza software for capturing horizontal positions was employed to empathize relative positional accuracy in an urban environment, during two seasons of the year, two times of twenty-four hour period, and two perceived WiFi usage periods. On average, time of year did not seem to influence the average fault observed in horizontal positions when GPS-only (no WiFi) adequacy was enabled, nor when WiFi was enabled. Observations of boilerplate horizontal position fault only seemed to improve with fourth dimension of twenty-four hour period (afternoon) during the leaf-off season. During each flavor and during each fourth dimension of day, horizontal position error seemed to improve in general during perceived high WiFi usage periods (when more than people were present). Overall average horizontal position accuracy of the iPhone 6 (7–13 thou) is consistent with the general accuracy levels observed of recreation-form GPS receivers in potential loftier multi-path environments.

Citation: Merry K, Bettinger P (2019) Smartphone GPS accuracy study in an urban environment. PLoS 1 14(seven): e0219890. https://doi.org/ten.1371/journal.pone.0219890

Editor: Filip Biljecki, National University of Singapore, SINGAPORE

Received: December 13, 2018; Accepted: July iv, 2019; Published: July 18, 2019

Copyright: © 2019 Merry, Bettinger. This is an open admission article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted apply, distribution, and reproduction in any medium, provided the original writer and source are credited.

Data Availability: All relevant data are within the manuscript and its Supporting Information files.

Funding: The writer(south) received no specific funding for this work.

Competing interests: The authors accept declared that no competing interests exist.

Introduction

Smartphones have get ubiquitous tools of the human race, as millions of people at present go nigh their days with small-scale GPS-capable computers in their hands or pockets. The majority of current research involving smartphone GPS capabilities focuses on transportation or directional uses [ane, ii, three, 4], patterns of homo movement [five, 6, 7], and health tracking [eight, nine, 10]. Two interests of lodge involve (a) whether the accurateness is sufficient to enable them to be a reasonable substitute for more expensive commercially available mapping devices, and similarly (b) to what grade of receiver (consumer or mapping) could smartphones be a reasonable substitute. Much of what we do (navigate) or produce (map) is contingent on the level of horizontal position accuracy under weather which GPS data is beingness collected. For example, a smartphone may not be the all-time pick for collecting mappable information in predominantly forested conditions, just may be reliable enough for data collection purposes in urban environments. While the expectations for GPS data quality using an iPhone should not be assumed comparable to the quality of data collected with a mapping-grade or survey-grade GPS receiver, hither we endeavour to appraise whether they can serve as a reasonable alternative and to measure what sort of error, both in distance and in direction, to expect. Therefore, the objective of this research was to assess the accuracy of the iPhone 6 GPS capabilities under various atmospheric condition. Specifically, this study is unique in the collection of static horizontal position data incorporating seasonal variability, level of human activity on the WiFi network, and urban woods conditions, and in assessing the part WiFi plays in GPS data drove accuracy.

Previous enquiry

Mod smartphones are equipped with Assisted GPS (A-GPS) capability. A-GPS uses smartphone networks in combination with a GPS antenna to increase the speed of determining or fixing position [11]. A-GPS precludes the need of a warm-up period required for traditional GPS units [12]. However, data nerveless using A-GPS is less accurate than traditional GPS receivers [13, 14, xv]. Location based services (LBS) allow 1 to access spatial positioning on a phone via cellular networks [16]. Depending on the user'south phone settings, LBS are activated when permitted applications (apps) are in use (east.g., while using map apps). With the convenience and ease of use of the GPS capabilities of smartphone devices, it seems of import to appraise their accuracy, so when used for data collection purposes the relative accuracy of positions determined tin can be understood. At that place are numerous apps bachelor to assist with GPS information collection. These permit the user to collect waypoints by simply tapping the screen of the phone. Depending on the app, a person can add their own basemap, interact with imagery provided past the app developer, and export saved positions from the app for utilize in other spatial software packages. A smartphone used in combination with an appropriate app can provide similar functionality as a bones recreation-grade GPS unit (due east.one thousand., Garmin, Magellan, etc.).

Mixed results tin exist found in the literature concerning the accurateness of smartphone GPS services. In a rather early report, using GPS enabled iPhones, iPods, an iPad and an app used by an insurance company in Switzerland, von Watzdorf and Michahelles [17] found average accurateness of location information between 108 and 655 m. More recently, the accuracy of static horizontal positions captured past a GPS-enabled phone was found to be around 20 1000 in one study [18]. This level of accuracy is well-nigh often influenced by landscape characteristics and the number of available satellites commonly leading to multipath errors. Multipath errors are the result of satellite signals bouncing off mural features like buildings, trees, or the ground before entering the device. In another instance, Menard et al. [nineteen] institute, in testing GPS accuracy across three dissimilar smartphone brands, that iPhone 4 determined approximately 98% of its GPS points inside x m of truthful positions and approximately 59% within 5 m. When admission to a WiFi network is available, that network is composed of admission points that are used to help identify location [14], as a WiFi admission point tin can emit its signal hundreds of meters. All the same, the number of WiFi access points available may accept no bear on on positional accurateness [20], contradicting early on research that indicated accuracy might improve with increased access signal availability [21]. Miluzzo et al. [22] conducted GPS information collection on the campus of Dartmouth College and determined that with a deterioration of smartphone service coverage and WiFi accessibility, accuracy also declined.

In add-on to the previously-noted studies, Zandbergen [xiv] found the average horizontal position error of an Apple tree iPhone 3G to exist around ten m, and Garnett and Stewart [23] establish the average fault for GPS points collected with Apple iPhone 4S to be around half-dozen.5 m. Similar to the methodology presented in our report, Garnett and Stewart [23] sought to decide whether time of day impacted positional accuracy. Using 3 collection periods, early morning, mid-day, and late afternoon, they plant that the kickoff and second collection periods had no significant impact on horizontal position accurateness. They likewise noted no touch on of weather weather condition on positional accurateness. While the accurateness was higher for the Garmin GPS units, the iPhone was comparable overall in open areas and areas with lower building heights. Through the incorporation of a differential correction method to data collected on an Android smartphone Yoon et al. [24] reduced positional error downwards to one k during both static and dynamic data collection. In comparison a Garmin GPSMap 66 and an Android phone, Lachapell et al. [25] tested the GPS information collection capabilities nether several unlike conditions including on a rooftop of a building, an urban canyon, indoors, and in a auto. They constitute a reduction in multipath issues using the GPS receiver compared to the smartphone. The Garmin GPSMap 66 had a root hateful square error (RMSE) beneath 1 m. Data collection in the vehicle using the Garmin unit again resulted in reduced multipath error. Modsching et al. [26] also noted that the presence of multi-story buildings tin can decrease the accuracy of horizontal adamant positions, due to use of degraded signals in club to generate position fixes inside urban canyons. Using a HTC G1 Dream and a Trimble Juno SB, Klimaszewski-Patterson [27] also assessed differences in accuracy between a smartphone and traditional GPS receiver. Using two different apps for GPS data drove, the residual error was lower when using the smartphone.

In contrast to the study of smartphones, more research has been conducted on the accuracy of traditional GPS receivers nether forested atmospheric condition. Wing et al. [28] investigated the accurateness of several consumer-form receivers across different forest canopy conditions (closed awning dumbo forest, twoscore–fifty% canopy cover of a young Douglas-fir (Pseudotsuga menziesii) stand up, and an surface area with no canopy cover). The average mistake for GPS collection ranged between 1 and four m in open areas, betwixt approximately i and 7 grand nether a moderately dumbo canopy, and approximately 3 and 11 yard under a closed forest canopy. In each forest condition, the average fault was less than the industry's assessment of 15 to 20 k accuracy. Within a dense Douglas-fir and western hemlock (Tsuga heterophylla) woods in Oregon, Wing and Eklund [29], measured comparable average positional accuracy between v and 9 g using a mapping-course GPS receiver and between 5 and 12 chiliad accurateness using a consumer-grade receiver following differential correction. Over a year's worth of information collection, Bettinger and Fei [30] evaluated the accuracy of a Garmin Oregon 300 GPS receiver. Information collection occurred nether varying woods conditions, within a hardwood stand and two pine stands with differing historic period classes, and nearly daily over the grade of a year. Horizontal position accurateness (about 6–11 m) was not impacted by environmental conditions similar temperature and relative humidity only varied significantly across forest stand types.

In a mixed deciduous-coniferous wood, Tomaštík et al. [31] evaluated horizontal positional accuracy of three cellphones (ZTE Blade, LG G2, and Sony M4 Aqua), a tablet (Lenovo Yoga 8), a survey-class GPS receiver and mapping-grade GPS receiver. Over the grade of a leaf-on and leaf-off menses, GPS data was collected at 74 points in a forested area at varying ages and density. Additionally, seventeen points were collected in an open meadow. For the three smartphones, average horizontal positional error ranged from half-dozen.74 to 11.45 m in leafage-on atmospheric condition, from 4.51 to 6.72 m in leaf-off conditions, and from i.90 to 2.36 in the open meadow.

Materials and methods

Study area

The University of Georgia, founded in 1785, is located in Athens, Georgia (Usa). The nigh 800-acre campus serves as an educational facility for approximately 40,000 students. PAWS-Secure is the proper name of the WiFi network attainable to students, faculty, and staff, and this organization is managed past the University's Section of Enterprise Data Technology Services. The WiFi network is accessible in all buildings and green spaces on campus. The University of Georgia as well has a system of survey monuments, benchmarks placed to identify surveyed points. Dispersed across the campus, the positions of these were surveyed and measured by the University Architects. The organization contains 212 survey monuments of which six were selected for utilise in this report (Fig 1). Of the vi monuments, five (Points 1–5) were established past surveyors in May 2015 and 1 was established in October 2003 (Point 6).

In an endeavor to describe the landscape surrounding each survey cap, a zone around each survey monument was created using a 30 m buffering process in a geographic information system (GIS). This 30 m buffer was chosen to represent an expanse twice equally big as the largest mean fault (14.fifteen m) from information collection beyond all survey monuments. A 2018 campus wide tree inventory provided past the University of Georgia's Warnell School of Forestry and Natural Resources [32], was used to describe the number of copse that fell within the buffer around each survey monument using a bespeak-in-polygon routine within GIS. A general description of the proximity (distance and direction) of each monument to nearby buildings was developed. Land embrace within the buffer was delineated using USDA NAIP imagery (1 one thousand spatial resolution) collected past the U.S. Section of Agriculture Farm Service Agency [33]. NAIP imagery is a digital ortho quarter quad (DOQQ) natural colour image collected during the growing season across the continental U.S. At the time of this enquiry, the well-nigh recently available imagery was nerveless in 2017.

Point (monument) ane was located in the center of a quad, with buildings on all sides and a network of sidewalks and grassy areas throughout. From Point one, there is a three story building approximately xl grand to the northeast, approximately 60 thou to the southeast is some other three story building, approximately twoscore m to the w is a four story building, and approximately 50 m to the southwest is a 2d four story building. Within the 30 m buffer, are thirty-nine trees that range in summit betwixt 2 and 22 grand. Tree species within the buffer include fringe tree (Chionanthus virginicus), southern magnolia (Magnolia grandiflora), northern white cedar (Thuja occidentalis), tulip poplar (Liriodendron tulipifera), Chinese elm (Ulmus parviflora), willow oak (Quercus phellos), and scarlet maple (Acer rubrum). Land comprehend within the 30 m was dominated by trees (xxx%), grass (37%), and sidewalks (22%).

Point 2 was located in a relatively open area, at the convergence of multiple walkways and adjacent to a route. The land embrace of the polygon buffer is 40% sidewalk and roadway, 28% tree cover, and 31% grass. From Bespeak 2, there is a iii story building approximately 35 m to the northeast from the monument. Around l one thousand from the monument to the northwest is a two story building, and 45 m to the southeast is a three story edifice. Within the 30 m sample buffer, there are thirty-iii trees ranging in pinnacle from 2 to 23 grand tall. The tree species included shortleaf pine (Pinus echinata), eastern redcedar (Juniperus virginiana), flowering dogwood (Cornus florida), water oak (Quercus nigra), willow oak, swamp anecdote oak (Quercus michauxii), cherry maple, and eastern redbud (Cercis canadensis).

Point 3 was located at the intersection of ii streets and approximately 30 grand from a three story edifice, the monument is located in shut proximity (approximately 3 k) to an 18 grand tall overcup oak (Quercus lyrata). An boosted 24 thousand tall overcup oak is approximately xv grand from the monument. In total at that place are eight copse within the 30 one thousand sample buffer. Other species include nuttall oak (Quercus nutallii) and pin oak (Quercus palustris). Heights of the trees here range from 2 thousand to 14 chiliad. The monument is also approximately 120 thou from the university's basketball game coliseum. The country comprehend near the survey marker is dominated by a combination of sidewalks, roadways, and a parking lot (44%). Nearly one-third of the buffer contained tree cover (29%) and low vegetation like small shrubs (24%).

Point 4 was located at an intersection of two streets. The dominant state cover surrounding the survey monument is comprised of sidewalk and roadway (58%). Both grassy areas (22%) and tree comprehend (16%) were in close proximity to the sample signal forth with a minimal corporeality of depression vegetation (iv%). In that location were two trees, alive oak (Quercus virginiana) and Yoshino ruby-red (Prunus x yedoensis), within the thirty thou sample buffer around this monument ranging in meridian betwixt 6 m and fourteen m. Approximately 40 k to the north of the monument is a three story building. To the southeast, approximately 35 m from the monument, is a ii story building.

Indicate v was located in a sidewalk next to a street. Inside the 30 m buffer around this monument, there are seventeen trees ranging in elevation from ii to 30 thou. The monument is located inside 10 m of the largest tree in the sample with the tree crown extending over the monument during the foliage-on menstruation. Tree species effectually this monument include pin oak, Nellie Stevens holly (Ilex 'Nellie R. Stevens'), trident maple (Acer buercerianum), Texas redbud (Cercis texensis), Okame red (Prunus x incamp 'Okame'), laurel oak (Quercus laurifolia), water oak, and pin oak. The majority of the land cover near the point was tree embrace (47%) in addition to grassy areas (22%), sidewalks and roadways (17%), and a pocket-size corporeality of low vegetation (vi%), and building (8%).

Point 6 was located inside a parking lot median. Only 6% of the area in proximity to the survey monument is comprised of low vegetation and grassy areas. The predominant land comprehend is sidewalks and roadways (52%) and tree cover (36%). The closest building to the monument is approximately 30 m away, and the building is 3 stories tall. Within the 30 m sample buffer around this monument, there are fifty-seven trees, including eastern red cedars, 2 willow oaks, eighteen flowering dogwoods, slash pine (Pinus elliotti), willow oak, tulip poplar, European hornbeam (Carpinus betulus), and a fringe tree, which range in height from 2 to 20 k.

Sampling design

A static horizontal position was recorded at each survey cap during 1 trip using LBS and the Avenza Maps app (https://www.avenza.com/avenza-maps/) on an iPhone 6. One trip entailed visiting each of the half dozen points one time in a clockwise or counter-clockwise order, as opposed to zig-zagging from point to betoken. For example, if a trip began at Point ane the side by side bespeak visited would either be Indicate 4 or Point 2. Similarly, if the starting point for a trip was Point 5, the next point visited would be either Signal half-dozen or Bespeak 3. In order to randomize the sampling procedure, the starting bespeak of each trip was randomized along with the direction of travel (clockwise or counter-clockwise). Trips were likewise separated by the time of day for sample drove. Samples were classified every bit morn (AM) samples collected between 8:00 AM EST and 11:59 AM eastern (USA) standard time, and afternoon (PM) samples collected betwixt 12:00 PM EST and 5:00 PM. The timing of trips was divided into high and low activeness times with respect to homo action on campus. High activity included days when classes were in session. Low activeness included days when classes were not in session (weekends, holidays, spring pause) or during the summertime when the number of students taking classes was quite lower. Every attempt was made to avoid collecting information when major sporting events occurred during what would unremarkably exist classified equally depression activity sampling periods. Finally, samples were nerveless during leaf-on and leaf-off periods. All foliage-on GPS collection was conducted between May 2017 and November 2017. All leafage-off GPS collection was conducted during two time periods: 1) from December 2016 to March 2017 and 2) March 2018. For the second time period, all data collection was completed during the month of March. During each data collection flow, each bespeak was visited 20 times. In full, there were eight split up data collection periods:

1. Period i: Loftier AM Leafage-on
2. Period ii: High PM Leafage-on
3. Period iii: Low AM Leafage-on
4. Period iv: Depression PM Leaf-on
5. Menses 5: High AM Leaf-off
6. Period 6: High PM Leafage-off
7. Period 7: Low AM Leaf-off
8. Period 8: Depression PM Leaf-off

Using a unipod and a level, the phone was positioned over each survey cap, with the data collector oriented to face due north (Fig 2). The phone was approximately three feet above the basis and was held away from the data collector's body during the information collection process. A cake of forest was placed on the top of the unipod to help in maintaining consistent positioning of the phone during data drove. Because of the blueprint of the phone and in an effort to position the GPS and WiFi antennas every bit closely as possible to the true surveyed position during the data collection process, the phone was held in two unlike positions. At the top of the telephone is both a 5 GHz antenna along with 2 GHz GPS / WiFi antenna. At the lesser of the phone is an additional WiFi antenna. When GPS-only sampling was conducted, the telephone was held horizontally with the height half of the phone positioned over the survey monument. When WiFi sampling was performed, nearly 3-quarters of the phone position was adjusted so that the eye of the phone was over the monument, since the 2 WiFi antennas are located on either end of the phone. For consistency, a slice of electric record was placed on the back of the phone approximately five cm from the top of the phone to mark where the phone should be positioned for GPS-only collection and some other mark approximately iii cm from the bottom of the phone for WiFi sampling. At each betoken, two GPS points were nerveless. For the first bespeak collected, the phone's WiFi capability was disabled. Afterwards the collection of the commencement point, the WiFi was enabled and 2 minutes were allowed to pass before the 2nd data bespeak was collected.

A total of 160 trips were completed effectually the examination class. During the leaf-on menstruation, 476 total horizontal positions were nerveless. There were 4 instances where, during a trip, data collection was non possible at Points 1 and iv due to structure vehicles parked over the monuments. During the leaf-off period, there was one instance where a data point was not nerveless at Bespeak 1 due to a vehicle being parked over a monument, therefore 479 information points were nerveless during the foliage-off period. During the leaf-on period, there were 29 instances where the WiFi adequacy was enabled for WiFi data collection but no WiFi connection was observed during the two-infinitesimal period post-obit GPS-only point drove. Whether a WiFi connection was fabricated was based on whether or not the WiFi status icon was activated on the telephone interface. In keeping with the pre-adamant data drove protocol, a waypoint was collected and a note was made that no WiFi connection had been made. Similarly, during the leaf-off data drove period, there were 38 instances where a WiFi connection was not evident. Beyond both leafage-on and foliage-off data drove periods, this state of affairs most oft occurred at Point half dozen, where 11 visits during leaf-off and 18 visits during leaf-on the WiFi status icon on the phone screen was not activated.

After the completion of each trip, the horizontal positions nerveless were exported from the phone as a .KML file so converted to bespeak shapefiles for use in GIS. Data were then converted from latitude / longitude to UTM NAD83 UTM Zone 17 N assuasive analysis to be conducted in the aforementioned units (meters) as the data associated with the monuments. Ane aim of this research endeavour was to determine whether using WiFi afflicted the horizontal position accurateness of the GPS data. Additionally, we were interested in whether an increased number of potential WiFi-users on campus had whatever impact on the horizontal position accuracy during data drove. Finally, we wanted to quantify the amount of horizontal position error a user could expect when collecting GPS points in an urban surroundings using an iPhone half-dozen. For this, Euclidean distances were determined between the monument position and the location of each horizontal position recorded.

Statistical tests

Statistical tests were employed to determine whether positional error from GPS-only and WiFi-enabled information were significantly dissimilar. Using nineteen collection categories (Table 1), the post-obit hypotheses were tested:

1. H_o: The horizontal position errors of the GPS-only and WiFi data are non distributed differently.
2. H_one: The horizontal position errors of the GPS-but and WiFi data are distributed differently.

The sets of static horizontal position errors of the nineteen dissimilar information collection scenarios were tested for normality using a Shapiro-Wilk test. We found that the sets of information were more often than not not ordinarily distributed. Therefore, a non-parametric test for statistical significance was required for statistical analysis [34]. Using the Mann-Whitney test is mutual when assessing GPS accuracy [35, 36, 37, 38, 39, xl]. The Mann-Whitney test was implemented at a 95% conviction level to examination the null hypothesis that the static horizontal position error nerveless during GPS-just data collection was non significantly different than the static horizontal position error collected when WiFi was enabled. Further, to describe the positional error, descriptive statistics including minimum and maximum error, and the RMSE of the horizontal position error were calculated for information collected at each sample point and under each of the 8 information collection periods. RMSE illustrates the error between the known location (the survey monument) and the nerveless location (waypoint). Specifically, RMSE is the square root of the average set of squared distances between a known location and the location recorded during data drove. RMSE is a common measurement of horizontal position accuracy in GPS enquiry [14, 30, 31, 41, 42]. In an attempt to identify what may be causing the horizontal position mistake, Pearson's correlation coefficient was used to determine whether country cover was correlated with the horizontal position accuracy of data. Pearson'southward correlation coefficient measures the clan between two variables expressed based on a value ranging between -1 (negative correlation), 0 (no correlation), and +one (positive correlation) [43].

Finally, local weather condition information variables were compiled for the menstruum of time during which the sampling bike was completed. Meteorological data was recorded for the time period of each trip using reports from nearby Athens Ben Epps Airport. Weather variables included air temperature (°F), relative humidity (%), barometric pressure (inches), wind speed (mph), and status (clear, partly cloudy, mostly cloudy, scattered clouds, overcast). Pearson'due south correlation coefficient was used to make up one's mind whether these conditions characteristics were correlated with the horizontal position accurateness of the data collected. Separately, a multivariate regression analysis was performed to identify the influence of atmospheric condition atmospheric condition on positional mistake. Regression analysis was chosen because it immune for an analytical method for incorporating categorical information. To exercise then, each weather condition condition was converted into a dummy variable containing values of 0 (the weather condition status was not recorded at the time of information collection) or 1 (the conditions was recorded at the time of data drove). Each conditions status served every bit an independent variable in the regression analysis with a total of five independent variables.

Results

In examining all horizontal position error derived during the GPS-only data drove effort, the minimum positional error was 0.05 grand compared to a maximum fault of 99.seven thousand. The RMSE for all data collected with the iPhone in GPS-just fashion was nearly nine.9 1000. On average, time of year did not seem to influence the average fault observed in horizontal positions when GPS-only capability was assumed. In terms of overall performance, some comeback was observed in RMSE during the leafage-off period, simply not in every case (e.g., High AM). Points iii and 5 seemed to have the highest RMSE during the leaf-on period, and were joined past Point 1 during the leaf-off period (Table 2). It should exist noted that relatively large confidence intervals were also derived from the sample information, indicating significant variability in the observed positional errors. The RMSE was generally lowest at Points 2 and 6, which were in relatively open areas, thus there is an assumed reduction in multipath fault. The overall average horizontal position fault was worst during the Low AM data collection period over both seasons. The single maximum horizontal position error observation was observed at Point 4 (virtually 100 grand), and is likely an outlier as the next largest single observation of horizontal position error was approximately 30 m (Table 3). The minimum horizontal position fault from a single ascertainment, beyond all points and seasons, cruel below 1 yard at least once in 20 of the 48 cases (6 points, 8 information collection periods) and roughshod below 2 g at least once in 39 of the 48 cases (Table 3).

On average, time of year also did not seem to influence the boilerplate error observed in horizontal positions when WiFi capability was enabled. Again, Points 3 and 5 seemed to have the highest RMSE during the leafage-on period, joined by Point i during the leaf-off period (Table 4). And every bit with the GPS-but information, relatively large confidence intervals were derived from the sample data at these points, indicating meaning variability in the observed positional errors. Farther, the RMSE was over again generally lowest at Points 2 and 6, which were in relatively open areas, thus there is an causeless reduction in multipath fault. In contrast to the GPS-simply results, the overall RMSE was highest during the High AM information collection periods during the leaf-off season. The single maximum horizontal position error ascertainment was observed at Point five (nearly 39 m) when WiFi was enabled (Table 5), and the single minimum horizontal position error observation was observed at Point half dozen (about 11 cm). The minimum horizontal position mistake from a single observation, across all points and seasons, vicious below 1 m in 15 of the 48 cases, and fell below two thou in 35 of the 48 cases (Table 5).

Across all of the combinations of data drove conditions examined, there were merely 12 (out of 133) instances where the sets of horizontal position fault were statistically significantly different (p < 0.05, Table 1). When considering all observations of horizontal position mistake during GPS-only data collection compared to all horizontal position mistake during WiFi information collection, the nada hypothesis was rejected and therefore the data are statistically significantly different (p < 0.05). However, nosotros could non reject the naught hypothesis when only considering the data nerveless from a single sample bespeak. The error was also statistically significantly different when considering all sets of observations of horizontal position error observed during AM, Leaf-off, and high data drove periods High AM and High PM (separately) data drove periods. Sample Betoken ane was the simply point where there were many instances of detection of pregnant differences among the GPS-only and WiFi-enabled data. While we cannot be completely certain what is causing the significant differences at this sample point, Point 1 is unique in that it is surrounded on each side by multistory buildings which may exist increasing the positional error when GPS-only data was collected. Nonetheless, each of these buildings firm WiFi access points which may lead to a stronger WiFi bespeak at this point reducing the positional fault when the WiFi is enabled.

In examining the frequency of the error across all points, interesting patterns sally (Fig 3). For instance, during GPS-but data collection, both Points two and vi had no instances of mistake sampling. Similar to Indicate 1, the occurrence of mistake was more often than not clustered in between 2 m and 20 thousand during GPS-only data drove. All the same, when WiFi was enabled, at that place were most twice as many occurrences of error (n = 76) between 5 and 10 m than whatsoever other error range. The mistake distributions at Bespeak 4 and Betoken 5 were very similar betwixt WiFi and GPS-but sampling with the majority of fault falling betwixt 2 to 10 yard at Bespeak iv and ranging from 2 g to more than xx m at Bespeak 5. The frequency of horizontal position mistake at Bespeak 6 was near prominent between the 0 and 2 chiliad and 2 to 5 m ranges for both GPS-only and WiFi data collection.

In an effort to identify what might be causing the horizontal error, a correlation between the percentage land encompass inside the xxx m sample buffer and the error was derived (Table vi). A moderate positive correlation between the building land cover class and horizontal positional error was plant during GPS-only data collection indicating that an increased presence of buildings led to an increase in horizontal positional error. This correlation was more pronounced during WiFi-enabled data collection but still just moderately. Minor positive correlations were found between tree cover and low vegetation and error during both GPS-only and WiFi-enabled data collection. Additionally, a small-scale negative correlation was found between the percent country cover classified as sidewalks, roads, and parking lots and positional error.

While understanding the amount of horizontal error betwixt a survey monument and a position collected by the iPhone 6 is useful, knowing the predominant direction of that error may also be important. In examining the directional error, the angle between survey monuments and positions adamant by the iPhone was calculated, and some general patterns emerged. For example, directional fault at Point 1 using GPS-only information predominantly ranged in cardinal management from s to due north with a bulk of directional mistake occurring in a west to northwest direction under all data collection conditions (Fig iv), nevertheless during data collection Period iii, there was no ascendant management of error. When WiFi was enabled, the direction of mistake was comparable to GPS-just (Fig 5) only well-nigh pronounced from the west to northwest. At Point two, at that place were several instances where in that location was no dominant management of fault during GPS-only data drove periods yet when because data collected under all data collection atmospheric condition the error almost often occurred from southwest to n. When WiFi was enabled, error typically occurred in a westerly pattern ranging from the due west-southwest to north-northwest. GPS-only information collection at Point three revealed directional error predominantly ranging from west-northwest to the northward-northeast. During two drove periods, Low AM leaf-on and Low PM leaf-on, at that place was no dominant directional error suggesting error was dispersed in all directions. During WiFi data collection, error during Menses one and Menstruum iii were consistently in a northerly direction clustered between due north-northwest and n-northeast, and in most other cases northwest to northeast. When WiFi was enabled at Point four, the direction of error often ranged between south-southwest to north-northwest while mistake was more dispersed across the fundamental directions for GPS-only data. During GPS-only data drove at Indicate 5, the vast majority of information error in all data collection conditions savage between the key directions of west-southwest and north. Similarly, when WiFi-enabled data collection occurred at Point 5, the bulk of management error brutal between west-southwest and north-northwest. Conversely, at Point vi when using GPS-only at that place were 5 different collection periods where there was no pronounced directional error, yet when directional error was pronounced, there was little consistency between data collection periods. When WiFi was enabled at this point, the directional error lacked a dominant direction.

Finally, well-nigh uniformly, there was generally weak to no correlation between horizontal position fault of the positions determined by the iPhone, and temperature, barometric pressure, air current speed, and relative humidity during the information collection try (Table 7). The results of the multivariate regression analysis indicated that there was no relationship between GPS-only positional error (adjusted R² = -0.0009) and atmospheric condition condition (i.eastward., clear, partly cloudy, etc). Similarly, when WiFi was enabled positional error was likewise not related (adapted Rⁱⁱ = -0005) to electric current weather conditions.

Discussion

When considering the static horizontal position error from all data points collected, the error observed in GPS-just information was significantly different from the error observed in the WiFi- enabled information. Farther, information collected in the morning, and information collected during high WiFi use periods also indicated that GPS-only data and WiFi-enabled data had significantly different levels of horizontal position fault. During simply the leaf-off season were like significant differences observed between GPS-merely and WiFi-enabled information. These observations, while non significant at every data collection point, advise that on average, enabling the iPhone to use WiFi signals to augment the determination of horizontal positions will lead to higher quality positional information. While information technology was unclear how extensive the WiFi services were utilized past the iPhone, the opportunity to use these services affected positional accurateness. Interestingly, nearby buildings may have influenced the management of error observed, due likely to multipathed signals from either the GPS satellite constellation or the WiFi signal emitting devices.

One blueprint became evident when interpreting the results for information nerveless when the WiFi was enabled: the average positional mistake was greater around Indicate v than all other test points, regardless of leaf-on or leaf-off conditions, and morning or afternoon data collection efforts, the error was virtually pronounced at this point. Some of this could likely be explained by multipath conditions or a uncomplicated deterioration in GPS signals. Yet, this survey monument was located across a two-lane street from a multistory hotel and convention center and under a large tree. Comparatively, Points 2 and half dozen were oft the data drove points with the lowest RMSE, specially during WiFi data collection periods. During GPS-only collection, Point 6 had low positional error compared to data collected at other survey monuments. Each of these monuments (Points ii and 6) were located in relatively open areas, and thus this may indicate that what plays the largest part in smartphone GPS data accuracy may be proximity to multistory structures, rather than increased activity on a nearby WiFi network or the presence of nearby copse. Further, our work has provided results that are similar to those provided by Garnett and Stewart [23], Weaver et al. [42] and others who have shown no correlation between atmospheric conditions and static horizontal position accuracy of low- to moderate-cost GPS receivers.

This observational report is i of the offset of its kind to examine the positional accurateness of horizontal positions determined past a smartphone during loftier and low human activity, and during two unlike seasons of the year (influencing the amount and presence of nearby tree canopies). Many of the influential factors could not be closely controlled by the written report squad; therefore, numerous samples were nerveless over each survey monument during random times of the mean solar day to understand on average the level of positional fault one might expect. While protocols for data drove seem reasonable (randomize the order of information collection, routinely collect data in the same manner at each monument, etc.), some of the uncontrollable aspects of the data collection process included the amount of human activity (high and low were all nosotros could presume), and the condition and condition of the WiFi network managed by the university. Persistent efforts were employed to larn information on the condition of the network during the information collection periods, yet we were unable to acquire metrics regarding the WiFi signal condition effectually the test course due to the university non assuasive u.s. access to these metrics. Equally a event, our observations should reflect average performance of the smartphone under average WiFi operating weather condition.

This study could exist complemented by further studies that focus on some of the limitations we observed. For instance, we were unable to sufficiently understand why horizontal position accuracy improved during periods of time when WiFi usage was high. Given our lack of access to the technical specifications of the WiFi network, perhaps management of the network during high use periods contributed to this result. Additional inquiry that incorporates a measurement of the strength of the WiFi signal at each sample point would be useful. Further, the results nosotros observed were highly variable around each sample point, perhaps due to the heterogeneous nature of the urban environment (spatial arrangement of trees, buildings, etc.). A complementary study to amend understand the bear upon of the spatial arrangement of features, similar to that of Bettinger and Merry [41] that was conducted in a forest, may farther our agreement of these issues. To further investigate the role multipath plays in error, information technology would exist useful to prepare a device at each sample point and continuously collect data over a flow of fourth dimension and at specified fourth dimension intervals. Hither, the assumption is that the error would repeat every bit long every bit the surrounding landscape (buildings, trees, etc.) remained the same. And finally, as computing applied science continues to evolve, continued observational and hypothesis-driven studies of smartphone accuracy in urban environments will be necessary to inform society of the potential practical and scientific uses of these hand-held positional and navigational devices. Specifically, a similar research endeavor using a newer smartphone with an improved GPS chip would exist valuable.

Conclusions

The horizontal position fault associated with GPS positions adamant past a smartphone is often assumed negligible by ordinary users of the engineering. Notwithstanding, as smartphones are used more often for data drove purposes, maybe during crowd sourcing data collection exercises or the capture of positional information through various smartphone apps, this concern may need more attention. Our study has shown that the overall boilerplate horizontal position error of the iPhone 6 is in the 7–13 chiliad range, depending on weather condition, which is consequent with the general accuracy levels observed of recreation-grade GPS receivers in potential loftier multipath environments. It seemed in our study that the time of twelvemonth did not influence the boilerplate horizontal position error observed when GPS-only parameters were assumed, or when WiFi was enabled. Our observations of average horizontal position mistake only seemed to improve with time of day (afternoon) during the leafage-off season. Interestingly, horizontal position mistake seemed to improve in general during perceived high WiFi usage periods (when more people were present) within each flavor and during each time of day most prominently in the afternoon. In general, directional fault was consistent at each data collection point during both GPS-only and WiFi drove. The nearly pronounced instance of directional error occurred at Signal v in a west to northwest direction. Data collection may take been subjected to multipath bug at some of the data collection points. We saw moderate correlation between the presence of buildings and positional error during both GPS-simply and WiFi-enabled data collection. Finally, weather conditions had picayune to no influence on the accuracy of information collected.

Supporting information

Acknowledgments

Nosotros would like to give thanks the Warnell Schoolhouse of Forestry and Natural resources for their continued support. We as well thank the anonymous reviewers for their time and effort.

References

1. 1. Das RD, Winter South. A fuzzy logic based transport way detection framework in urban environment. J Intelligent Trans Syst. 2018; 22: 478–489.
   • View Article
   • Google Scholar
2. 2. Nour A, Hellinga B, Casello J. Classification of automobile and transit trips from smartphone information: Enhancing accuracy using spatial statistics and GIS. J Transp Geogr. 2016; 51: 36–44.
   • View Commodity
   • Google Scholar
3. three. Hofer H, Retscher G. Seamless navigation using GNSS and Wi-Fi/IN with intelligent checkpoints. J Location Based Services. 2017; 11: 204–221.
   • View Article
   • Google Scholar
4. iv. Vlahogianni EI, Barmpounakis EN. Driving analytics using smartphones: Algorithms, comparisons and challenges. Transp Res Part C Emerg Technol. 2017; 79: 196–206.
   • View Article
   • Google Scholar
5. 5. Lue One thousand, Miller EJ. Estimating a Toronto pedestrian road selection style using smartphone GPS data. Travel Behav and Soc. 2019; xiv: 34–42.
   • View Article
   • Google Scholar
6. 6. Zhou 10, and Li D. Quantifying multi-dimensional attributes of human activities at various geographic scales based on smartphone tracking. Int J Health Geogr. 2018; 17: eleven. pmid:29743069
   • View Article
   • PubMed/NCBI
   • Google Scholar
7. vii. Glasgow ML, Rudra CB, Yoo EH, Demirbas M, Merriman J, Nayak P, et al. Using smartphones to collect time-activeness data for long-term personal level air pollution exposure. J Expo Sci Environ Epidemiol. 2016; 4: 356–364.
   • View Article
   • Google Scholar
8. 8. Obuchi SP, Tsuchiya S, Kawai H. Examination-retest reliability of daily life gait speed as measured by smartphone global positioning systems. Gait Posture. 2018; 61: 282–286. pmid:29413798
   • View Article
   • PubMed/NCBI
   • Google Scholar
9. 9. Hardy J, Veinot TC, Yan X, Berrocal VJ, Clarke P, Goodspeed R, et al. User acceptance of location-tracking technologies in wellness research: Implications for study design and information quality. J Biomed Inform. 2018; 79: vii–xix. pmid:29355784
   • View Article
   • PubMed/NCBI
   • Google Scholar
10. 10. Aranki D, Kurillo M, Yan PS, Liebovitz DM, Bajcsy R. Existent-time monitoring of patients with chronic middle-failure using a smartphone: Lessons learned. IEEE Trans Affect Comput. 2016; vii: 206–219.
    • View Article
    • Google Scholar
11. eleven. Vallina-Rodriguez N, Crowcroft J, Finamore A, Grunenberger Y, Papagiannaki Thousand. When help becomes dependence: characterizing the costs and inefficiencies of A-GPS. GetMobile. 2013; 17: 3–14.
    • View Commodity
    • Google Scholar
12. 12. Bierlaire M, Chen J, Newman J. A probabilistic map matching method for smartphone GPS data. Transp Res Function C Emerg Technol. 2013; 26: 78–98.
    • View Article
    • Google Scholar
13. 13. Massad I, Dalyot South. Towards the crowdsourcing of massive smartphone assisted-GPS sensor ground observations for the production to Digital Terrain Models. Sensors. 2018; 18: 898.
    • View Commodity
    • Google Scholar
14. fourteen. Zandbergen PA. Accuracy of iPhone locations: a comparing of assisted GPS, WiFi and cellular positioning. Trans GIS. 2009; 13: 5–26.
    • View Article
    • Google Scholar
15. xv. Zandbergen PA, Barbeau SJ. Positional accuracy of assisted GPS data from high-sensitivity GPS-enabled mobile phones. J Navig. 2011; 64: 381–399.
    • View Commodity
    • Google Scholar
16. 16. Bauer C. On the (in-)accurateness of GPS measures of smartphones: a written report of running tracking applications. Proceedings of International Conference of Advances in Mobile Computing and Multimedia. 2013, December. Vienna, Austria.
17. 17. von Watzdorf Southward, Michahelles F. Accuracy of positioning data on smartphones. Proceedings of the 3rd International Workshop of Location and the Spider web 2010. 2010, Nov. Tokyo, Japan.
18. xviii. Mok East, Retcher G, Wen C. Initial test on the use of GPS and sensor information on modern smartphones for vehicle tracking in dense loftier rise environments. Proceedings of the Ubiquitous Positioning, Indoor Navigation, and Location Based Services 2012. 2012, Oct. Helsinki, Finland.
19. 19. Menard T, Miller J, Mowak M, Norris D. Comparing the GPS capabilities of the Samsung Galaxy S, Motorola Droid 10, and the Apple iPhone for vehicle tracking using FreeSim_Mobile. 14th International IEEE Briefing of Intelligent Transportation Systems. 2011, October. Washington, DC.
20. 20. Zandbergen PA. Comparison of WiFi positioning on ii mobile devices. J Location Based Services. 2012; half-dozen: 35–50.
    • View Article
    • Google Scholar
21. 21. Wallbaum M. A priori error estimates for wireless local area network positioning systems. Pervasive and Mob Comput. 2007; 3: 560–580.
    • View Article
    • Google Scholar
22. 22. Miluzzo Eastward, Oakley JMH, Lu H, Lane ND, Peterson RA, Campbell AT. Evaluating the iPhone as a mobile platform for people-centric sensing applications. Proceedings of the International Workshop on Urban, Community, and Social Applications of Networked Sensing Systems. 2008, Nov. Raleigh, NC.
23. 23. Garnett R, Stewart R. Comparison of GPS units and mobile Apple tree GPS capabilities in an urban mural. Cartogr Geogr Inf Sci. 2015; 42: 1–viii.
    • View Article
    • Google Scholar
24. 24. Yoon D, Kee C, Seo J, Park B. Position accuracy improvement by implementing the DGNSS-CP algorithm in smartphones. Sensors. 2016; 16: 910.
    • View Article
    • Google Scholar
25. 25. Lachapell Thousand, Gratton P, Horrelt J, Lemieux Due east, Broumandan A. Evaluation of a low cost manus held unite with GNSS raw information capability and comparison with an Android smartphone. Sensors. 2018; 4185.
    • View Article
    • Google Scholar
26. 26. Modsching G, Kramer R, 10 Hagen Grand. Field trial on GPS accuracy in a medium size city: the influence of built-upward. Proceedings of the 3rd Workshop on Positioning, Navigation, and Communication (WPNC' 06). 2006, Mar. Hannover, Frg.
27. 27. Klimaszewski-Patterson A. Smartphones in the field: preliminary study comparing GPS capabilities between a smartphone and dedicated GPS device. Papers of the Applied Geography Conferences. 2010; 33: 270–279.
    • View Article
    • Google Scholar
28. 28. Wing MG, Eklund A, Kellogg LD. Consumer-grade global positioning (GPS) accuracy and reliability. J Forestr. 2005; 103: 169–173.
    • View Article
    • Google Scholar
29. 29. Fly MG, Eklund A. Performance comparison of a low-cost mapping form global positioning systems (GPS) receiver and consumer course GPS receiver under dense wood canopy. J Forestr. 2007; 105: 9–14.
    • View Article
    • Google Scholar
30. 30. Bettinger P, Fei Southward. One year'south experience with a recreation-grade GPS receiver. Mathematical and Computational Forestry & Natural-Resource Sciences. 2010; 2: 153–160.
    • View Article
    • Google Scholar
31. 31. Tomaštík J Jr, Tomaštík J Sr, Saloň Š, Piroh R. Horizontal accurateness and applicability of smartphone GNSS positioning in forests. Forestry. 2016; 90: 187–198.
    • View Article
    • Google Scholar
32. 32. Fob W. Estimating total carbon sequestered in trees of the University of Georgia. Poster presented at: Highlighting UGA's undergraduate research, 2018 CURO Symposium. 2018, April 9–10. Athens, Georgia.
33. 33. U.S. Department of Agriculture. Imagery programs, NAIP Imagery. USDA Subcontract Service Agency, Aerial Photography Field Part, Table salt Lake City, UT. 2011. Bachelor online at: https://world wide web.fsa.usda.gov/programs-and-services/aerial-photography/imagery-programs/naip-imagery/. Cited 21 February 2019.
34. 34. Sokal RR, Rohlf FJ. Biometry. third ed. New York: Westward.H. Freeman and Company; 1995.
35. 35. Ucar Z, Bettinger P, Weaver S, Merry KL, Faw Yard. Dynamic accuracy of recreation-class GPS receivers in oak-hickory forests. Forestry. 2014; 504–511.
    • View Commodity
    • Google Scholar
36. 36. Abdi E, Sisakht SR, Goushbor L, Soufi H. Accuracy assessment of GPS and surveying technique in forest road mapping. Register of Woods Inquiry. 2012; 309–317.
    • View Article
    • Google Scholar
37. 37. Rodríguez-Pérez J, Álvarez MF. Assessment of low-cost GPS receiver accuracy and precision in forest environments. Periodical of Surveying Technology. 2007; 159–167.
    • View Article
    • Google Scholar
38. 38. Moriarty KM, Epps CW. Retained satellite data influences performance of GPS devices in a woods ecosystem. Wildlife Society Bulletin. 2015; 349–357.
    • View Article
    • Google Scholar
39. 39. Barrette J, August P, Golet F. Accuracy cess of wetland boundary delineation using aerial photography and digital orthophotography. Photogrammetric Engineering & Remote Sensing. 2000; 409–416.
    • View Article
    • Google Scholar
40. 40. Zandbergen PA, Barbeau SJ Positional accuracy of assisted GPS data from loftier-sensitivity GPS-enabled mobile phones. The Journal of Navigation. 2011; 381–399.
    • View Article
    • Google Scholar
41. 41. Bettinger P, Merry K. Influence of the juxtaposition of trees on consumer-class GPS position quality. Mathematical and Computational Forestry & Natural-Resources Sciences. 2012; 4: 81–91.
    • View Article
    • Google Scholar
42. 42. Weaver SA, Ucar Z, Bettinger P, Merry K. How a GNSS receiver is held may touch on static horizontal position accurateness. PLoS Ane. 2015; 10: e0124696. pmid:25923667
    • View Article
    • PubMed/NCBI
    • Google Scholar
43. 43. McGrew JC Jr, Monroe CB. An Introduction to Statistical Problem Solving in Geography. 2d ed. United States: McGraw-Hill; 2000.

How Accurate Is Cell Phone Gps Location,

Source: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0219890

Posted by: holterfatert94.blogspot.com

Share this post