Effect of occupation time on the horizontal accuracy of a mapping-grade GNSS receiver under dense forest canopy

A mapping-grade dual frequency GNSS receiver was tested under dense forest canopy to determine the effect of occupation time on horizontal accuracy. The U.S. Forest Service Forest Inventory and Analysis unit in the Pacific Northwest has been using 32 of these units to collect over 7,000 plot locations since 2013. In this study, one-hour cNss static occupations were collected at 33 ground-surveyed positions with Trimble GeoXH6000 mapping-grade and Javad Triumphl surveygrade receivers. Rover files were differentially post-processed and horizontal accuracy of each post-processed position was computed. Results indicated that 1.85 m accuracy (n = 0so) could be achieved with the GeoXH6000 receiver with 1 5 -minute occupations ; however, maximum horizontal error was 7.01 m. Increasing occupation time to 20 minutes did not result in a significant improvement in accuracy. No correlation was found between the horizontal precision of a postprocessed position reported by the postprocessing software and the field-measured horizontal occuracy of the positions. lntroduction National Forest Inventories (NpI) are designed to produce and report estimates of forest resources. The types of estimates produced include forest area, volume, condition, removals, growth, mortality, and overall trends in Iand use. Estimates are generally summarized to represent various levels, including owner group (e.g., state, private, federalJ, ecological units, survey unit, forest type, and tree species. Inventory data have traditionally been used by public and private land managers, planners, and researchers to address a variety of information needs. NFIs are established in many countries around the world, providing information to the local government and commercial timber industry as well as supporting global efforts to monitor the effects of climate change, carbon emissions and sequestration, and biological diversity. In the United States, The Nru is conducted by the Forest Service's Forest Inventory and Analysis program (rn). na collects, analyzes, and reports information on the status and trends of America's forests: how much forest exists, where it exists, who owns it, and how it is changing, as well .as how the trees and other forest vegetation are growing and how much has died or has been removed in recent years. FIA uses four related surveys to characterize different aspects of America's forests: forest monitoring, ownership survey, timber product output survey, and utilization studies. The forest monitoring component uses a three-phase sample with permanent sample sites located Robert J. McGaughey, Hans-Erik Andersen, and Stephen E. Reutebuch (retired) are with the USDA Forest Service, Pacific Northwest Research Station, University of Washington, PO Box 3 5 2 100, Seattle, WA, 98 1 95-2 1 00 (bmcgaughey@fs.fed.us). Kamal Ahmed is with Cairo University, Faculty of Engineering, Cairo, Egypt across the United States. Phase 1 uses remotely-sensed data for stratification and to identify whether or not a location is forested. Phase 2 consists of one sample site for every 6,000 acres, where fie1d crews collect data on forest type, site attributes, tree species, tree size, and overall tree condition. Phase 3 utilizes a subset of the Phase 2 sample sites where crews measure a broader suite of forest health attributes including ttee ctown conditions, lichen community composition, understory vegetation, down woody debris, and soil attributes. rIa samples and summarizes data for the continental states, Alaska, Hawaii, and US Pacific Island Territories and Protectorates. The program includes over 323,000 phase 2 sample locations. Approximately 116,400 of these locations are forested and about 20,000 are measured each year (Vogt and Smith, 201,7).Ttre prR database is one of the most comprehensive and complete samples of vegetation conditions in the world. For the sample sites, coarse locations are obtained using aerial imagery to guide crews to the site. In the field, crews typically record a location using a global navigation satellite system (cNss). Plot confidentiality requirements limit the distribution of the actual plot locations. However, cooperators can obtain and use the actual locations after completing a formal spatial data request process and signing a non-disclosure agreement. For pn, the location (typically acquired using a low-cost, consumer-grade cps unit) is primarily used to relocate the sample site. While an accurate position is always desirable, the position only needs to be good enough to help crews plan for travel to and from the site and to relocate the site each time it is measured. Accuracies for the GNSS receivers iypically used by field crews range foom a few meters to several decameters depending on the quality of the cNSS receiver, the forest conditions at the site, and postprocessing of the cNss data (Bolstad et aL.,zoos; Hoppus and Lister, 2007J. This level of accuracy has been sufficient for pn. However, the extensive area covered by rra data, repeated measurements over time, consistent protocol, and general availability of the data make the plot data useful for a variety of applications many of which would benefit from more accurate locations. NFI data have been combined with a variety of remotely-sensing data to map characteristics over large areas and to help detect changes over time (Tomppo et al.,2oo\; Zald et a1.,201,4; Blackard et al., 2OOB). Historically, rectification error present in the remotely-sensed data (Landsat rll) has been such that improving the accuracy of the cNSS location did not always result in improvements in analysis results (McRoberts, 2010). However, the remote sensing world is changing. High resolution digital Photogrammetric Engineering & Remote Senslng Vol. 83, No. 12, December 2017, pp.861-868. oogs-'1LL2 I L7 I 861,-868 @ 201,7 American Society for Photogrammetry and Remote Sensing doi: 10.14358/PERS.83.12.861 PHOTOGRAMI,lETRIC ENGINEERING & RE14OTE SENSING II November 241/ 861 imagery and lidar typically have horizontal registration errors of less than 1 m. When considering the size of the p'ra subplot (circular plot with radius of 7 .3 m) , the effect of a 10 m emor in location can have a significant impact on the ability to describe relationships between plot measurements and remotelysensed attributes. For example, when developing relationships between lidar point cloud metrics and forest structure metrics, a piot position horizontal accuracy of 1, to 2 m ensures that ground measurements are reasonably well aligned with lidar data (Frazer et al.,2oLt; Gobakken and Nesset, 2009). Within FIA, GNSS receivers have been in common use for over 25 years. However, early receivers tracked few satellites (five channels) and did not allow diff'ereniial correction of positions (Hoppus and Lister, 2007). Over the last decade, crews have relied on recreational-grade receivers that use the wide area augmentation system (lvaasJ to improve the accuracy of positions. In 2013, FIA crews based in the Forest Service's Pacific Northwest Research Station (lNw-nm is responsible for measurements in Alaska, California, Oregon, Washington, Hawaii, and U.S. Pacific Island territories and Protectorates) began using mapping grade cNss receivers (Trimble GeoXH 6000 series standard edition with decimeter accuracy) that allow differential postprocessing of the collected rover flles. As of 2016, 32 of the Trimble units were in use in pNw-pta and locations had been collected for over 7,000 plots using the Tiimble units. While these practices are intended to improve plot locations, even differentiaily-corrected static coordinates coilected under forest canopy can have horizontal errors ranging from sub-meter to tens of meters, depending on many factors (e.g., density of canopy, number and geometry of available satellites, duration of data collection, type of receiver and antenna, Iocal topography, availability ofreliable base station data, and characteristics of differential processing software). Large errors often occur even when using survey-grade dualfrequencv, multi-constellation receivers with moderately long occupation times (1s-min to 1-hour) and with data from 10 or more satellites (Clarkin, 2007; Andersen et aL,2009; Valbuena et ol., 2o1.o). Nesset (zoor) tested a dual-frequency cNSS (cPS and cLoNass) receiver under typical Scandinavian forest conditions and reported horizontal errors from 0.08 to 1.35 m with 2.5 to 20 min static occupations. Edson and Wing (2012) tested a mapping-grade cPs-only receiver in relatively low density conifer stands in Oregon with 30 second occupations and reported errors in the 2 to 3 m range. Frank and Wing (2014) tested a mapping-grade cPs-only receiver in low to high density conifer stands in Oregon with 1, 5, 10, 30, and 60 second occupations and reported errors in the 4 to I m range. Large errors occur because ofphenomena known as signal multipathing and non-1ine-of-sight (Nrosl reception when using cttss receivers in forested environments. A multipath signal occurs when the direct line of sight signal from a cNSS satellite and a reflected signal from the same satellite are received by the GNSS receiver. In forested areas, the direct line-of-sight signal between a cNSS satellite and the ct'tss receiver can be temporarily blocked by an object such as a tree bole. When this occurs, only a reflected signal (wlos) may be received. The ranging distance for a NLos signal is greater than the direct line-of-sight range from the satellite to the receiver; whereas, a multipath signal can either increase or decrease the range. If the multipath or Nt-os signal is not detected by the receiver or postprocessing software the range can cause large errors in the computed position. Petovello (2013) provides a detailed discussion of multipath and NLos signais and their effects on GNSS positioning. Because the navigation satellites are not geostationary, the line of sight from the receiver to each