Additional Details of GPS Machine Tracking Work
Appendix 2. Additional Details of GPS Machine Tracking Work
The Sokkia Spectrum unit is a handheld differential GPS consisting of a separate receiver, data logger, battery and external antenna. The antenna was mounted on the roof of the machine cab. A steel protector was fabricated to protect the aerial from limbs and debris while not hindering the antenna’s ability to receive satellite signals. The GPS receiver, data recorder and battery were placed in a cloth bag, reinforced with additional padding, and housed within the cab. A sealed lead-acid battery was used to power the GPS unit. The unit was set to record data in stream mode at a five second interval with a 15-degree elevation mask. Recording at a five-second interval, the GPS unit stored data for approximately 10 hrs. The data logger, receiver and battery were removed from the machine daily to download the data and recharge the battery.
The CSIRO FFP forest machinery data logger is one of a series of three identical units constructed by the Forestry Operations research team at CSIRO for machine monitoring. It is a small robust computer system assembled from electronics cards fabricated to the PC104 standard. The system is functionally equivalent to an IBM PC Model AT. A GPS card supplies location information, and up to 50 Mb of data are stored on removable Compact Flash cards. The assembled computer, GPS and data storage cards are housed in a robust aluminium housing about 150mm on each side (Fig 1.). The system was powered though an electrical circuit that received power only while the engine was operating. The data logger began recording position approximately 30 seconds after machine start-up and operated continuously until machine shutdown. Data were collected at five- second intervals. A small magnetically mounted aerial was used for signal capture and it was attached to the rear corner of the skidder roof. The aerial was unguarded. The Flash Card storage chips had sufficient capacity to record about 10 days work but were changed weekly.
The GPS data were subsequently differentially corrected using base station data collected in Canberra in the Australian Capital Territory (ACT) for the NSW study and Melbourne for the Victorian study. Canberra is located 200 km from the field sites in Eden and the Melbourne base station was about 100 km from the field site. There were several periods for which differential processing was not possible, due to equipment failure in both the field and at the base stations. For these periods, data accuracy was reduced and traffic patterns had to be analysed manually, as described below.
Importing GPS Data The GPS data files contained fields for latitude, longitude and height for each point. A typical
stream file for a single day of data collection could contain five thousand or more points. There were some “gaps” in the GPS data where the satellite signal was lost due to poor satellite position, too few satellites, or lost signal due to topography and canopy cover. To account for these gaps in the GPS data an ArcView GIS program was used to filter the data during the importing stage. It flagged missing points and evaluated the distance travelled in each observation interval. Errors such as those induced by multi-path signal reception cause major jumps in reported position (up to 500 metres was observed). By setting a maximum plausible travel distance for the five-second interval, observations could be processed to identify and flag occurrences of implausible jumps. These were typically followed in the next observation by a return jump to the correct position. Data were corrected after visual inspection, by deleting the erroneous point and assuming the travel path was a straight line between the point before and the point after the jump.
There were large differences in the apparent data quality for GPS derived machine position between the two trials (NSW and Victoria). The Victorian data set included numerous periods of lost signal or unstable position, identified by the GIS data filter and readily evident in screen plots of daily operating pattern. This is evidenced as both location jumps and widely spaced but parallel travel path recordings. As a check, field inspection was employed to identify actual track location There were large differences in the apparent data quality for GPS derived machine position between the two trials (NSW and Victoria). The Victorian data set included numerous periods of lost signal or unstable position, identified by the GIS data filter and readily evident in screen plots of daily operating pattern. This is evidenced as both location jumps and widely spaced but parallel travel path recordings. As a check, field inspection was employed to identify actual track location
Map Production Traffic maps were produced in two ways. The preferred method used the automated tool set
developed and installed in the ARCVIEW GIS. The back up method relied on visual inspection of the GIS data on screen, counting and manual recording of traffic path. This was used when automated processing was not possible.
In the automated method, a composite GIS file containing all snigs on the site was assembled first. From this a simplified schematic of the snig track network was drawn as an overlay layer in the GIS, identifying the track sections and junctions. Next, a set of lines were drawn on the overlay layer at each point where traffic count was desired. These counting lines were drawn to intersect the pattern of the snig track sections at right angles, cutting directly across the snig track, usually midway between junctions. Figure 2 shows a section of track, associated data and counter bars. The automated computer routine was programmed to count the number of GPS track records that intersect each of these counting lines and report the results.
Some parts of the data set could not be processed automatically, either where there were numerous jumps or where inspection of the daily work paths showed some parallel track records to be too widely spaced. Widely spaced track records prevent the analyst drawing counting bars that were mutually exclusive between nearby physical tracks. In these cases, counting was done by individually tracing the machine path from the ARCVIEW records, and recording passes manually on the simplified snig track map. Automated processing was possible for about two-thirds of the NSW data. Manual processing was undertaken on the Victorian data and the remaining parts of the NSW data.
Data quality was observably poorer (parallel tracking, jumps) for several periods during the NSW trial and for longer periods in the Victorian trial. The most significant effect was in greatly increasing the time needed to process the traffic data due to the requirement for manual processing. Comparatively lower levels of accuracy under forest canopy cover are a well-recognised phenomena in forestry GPS traffic studies. Accuracy is primarily determined by the class of GPS receiver used and signal quality. Accuracies of about 2 metres with 95% probability are commonly reported for differentially corrected data using this class of receiver. Previous studies using the same Sokkia equipment under similar eucalyptus canopies produced accuracies of between 2 and 4 metres (McCormack et al, 2000). Common reasons for the reduction in performance under canopy are blocking or reflection of signal from the satellite. The GPS system relies on a direct signal path from satellite to receiver. The signal strength at the earth surface is very low, and easily blocked by tree canopy. A related difficulty, which may have contributed to the poorer result in Victoria is signal reflection (termed multi-path), for example, from the trunks of nearby trees at the clearfall edge. This problem is widely reported in GPS studies. Finally, there may have been difficulties related to the type of aerial and its positioning and functioning on the skidder in Victoria where a smaller magnetic mount aerial was used. Higher levels of debris build-up on the cab roof were noticed in the much higher yielding Mountain Ash forest and this may have periodically interfered with signal reception.
The statistic evaluated in this study was the number of machine passes along the snig track. At this level of accuracy it is not possible to have confidence about the accuracy of the actual number of wheel passes over a point. Analysis was, therefore, also undertaken to condense the data to broad traffic classes (eg. 1-10, 11-25 25+ etc).
Figure 1. CSIRO Forest Machine Data Logger