Chapter 7 Nature of Spatial Data, Accuracy, and Validation
7.5 Case Study of Hong Kong
Road collision data in Hong Kong, as many cities and countries worldwide, are pri- marily collected by the police. Specifically, the police filled out the “Traffic Accident Report Booklet (Injury Case)” for traffic collisions involving injuries. The data are processed and computerized by the police and the Transport Department into the collision, vehicle, and casualty databases. These three databases represent the major sources of road collision information in Hong Kong and are collectively known as TRADS. At present, only the collision database of TRADS contains spatial variables.
In particular, the five-figure grid references (GRID_N, GRID_E) give the precise location of a road collision. Apart from the grid references, three additional codified locational variables—first street name (ST_NM), second street name (SECND_ST), and district board (DBOARD)—and three textual variables that describe the nearby landmarks (IDEN_FTR), precise locations (PREC_LOCTN), and circumstances (HAPPEN) in the TRADS collision database are found.
Next, the geovalidation involves the building of a link-node system about the road network in Hong Kong. In this respect, the Lands Department maintains the digitalized road network database in Hong Kong. As explained earlier, the creation of standardized buffer zones creates double-counting and false error problems,
Delete the original record Next record of road accident
Road accident records Yes Yes
Yes Correct Correct
Correct CorrectIncorrect
Incorrect Incorrect
Incorrect
No No
No
Has the record been checked? Save the amended record attributesCheck whether the accident happened at junction Check the road names with the network database
Snap to the nearest junctionSnap to the nearest road Check district of the accident record with the district board database Assign to the correct district board
Intersect road accident location with the district
board database
Save the new location and copy the accident attributes
Check district of the accident record with the district board database Insert the road
accident location with district board database
Check the road names with the road network database Match the amended road accident record’s road names with the road network database
Manually check for error and correction
Match foundNot found
Snap the next nearest point by (1) Removing the road/junction of the incorrec
t snapped point in the road network database; and then (2) Snapping to the next nearest road/junction of the road network database
Since the next nearest point does not yield better results, (1) Revert to the nearest road/junction of the road network database (2) Check whether road name is missing in one of the database
s
Assign to the correct district board Key Data Decision Process
FIGURE 7.3The GIS-based spatial data validation system. (Reprinted from Acc. Anal. Prev., 38(5), Loo, B.P.Y. Validating crash locations for quan- titative spatial analysis: A GIS-based approach, 879–886, Copyright 2006, with permission from Elsevier.)
Nature of Spatial Data, Accuracy, and Validation 143
especially in areas with narrow and dense roads (like urban Hong Kong) and near junctions. The Cartesian coordinates (X and Y) and the road name (ST_ENGNM) in the road network database are later used for matching information in the collision database.
Finally, this spatial data validation system makes use of the district board cover- age. District board is chosen as the spatial subunit for validation, because it is the territorial unit not only for local election but also for the compilation and release of most official demographic and socioeconomic statistics in Hong Kong. Locational information of the collision database of TRADS is later validated against the spatial variables (X, Y, and Dist_Abbre) of the district database.
7.5.2 methodology
Following the above conceptual framework of Figure 7.3, a six-stage GIS-based spa- tial data validation system is developed. In Loo (2006), the validation results of the police-recorded road collisions in Hong Kong from 1993 to 2004 were presented.
In this chapter, updated validation results from 2005 to 2010 are shown in Table 7.1 to illustrate the methodology. Over this period, the total number of police-recorded road collisions in Hong Kong has stayed at around 15,000 per year.
In the first step, road collisions are snapped to the link-node system. To see whether the road collisions intersect with the link-node system, the five-figure grid references are first transformed into a GIS-compatible format (by adding the prefix
“8” in this case). Then, a simple intersection process was performed in GIS. To distinguish between collisions happening at junctions and mid-block locations, the information contained in the textual descriptive variables is utilized. Specifically, a standardized library containing the terms used for denoting collisions happening at intersections is compiled. Examples of these terms used in Hong Kong include
“intersection,” “junction,” “J/,” and “JW.” If the spatial variables contain these key- words, the collisions are snapped to the nearest junctions on the link-node system.
Otherwise, the collisions are snapped to the nearest centerline of the road network.
From 2005 to 2010, about 40%–45% of the collisions intersected with the link-node system. The shares were much higher than the 1993–2004 period with less than 1%
of the collisions (Loo 2006). After snapping the collision locations to the nearest points on the link-node system, it was found that about 90%–95% of the road colli- sions could be validated as having both correct road names and district boards. Once again, there have been substantial improvements, when compared with 79.2% back in 2004 (Loo 2006).
At the second step, road names of the collision records are matched with the road network database. Since the road names in Hong Kong do not follow any particular system (e.g., the numbering system from north to south or east to west), another library was developed to identify road names in the textual spatial variables.
Examples of these terms include “Road,” “Street,” “Avenue,” “Path,” “Circuit,”
“Highway,” “Roundabout,” “Lane,” and their numerous forms (including abbrevia- tions). If the road name matching is successful, the spatial variables of the collision database are then validated against the district board database in the third step. If not, the fourth step is triggered, that is, collisions are snapped to the next nearest
road junction or section before a new round of road name matching exercise. If the matching is successful, the district board information is then validated again. This step is necessary because the next nearest point may lie in a different district board.
During 2005–2010, about 3.5%–4.5% of the collision records were found to be hav- ing correct road names but wrong district board information. Should the focus of validation be put on road names only, 88.2%–90.7% of the collision records could be considered as correct. After step four, the cumulative percentage of validated colli- sion records increased to 97.7%–98.2% in 2005–2010.
For unsuccessful matching at the second round, the fifth step of the system is to identify the road names recorded in the spatial variables of the collision database and then try to find a match in the road network database. This step can be performed by the address matching function of GIS (see, e.g., Levine and Kim 1998). If the match- ing is successful, the collision is snapped to the nearest road junction or section of TABLE 7.1
R esults of the Geovalidation of Traffic Collisions in Hong Kong, 2005–2010
2005 2006 2007 2008 2009 2010 The raw collision database
Number of road collisions 15,062 14,849 15,315 14,576 14,316 14,943
On road centerline (%) 44.5 43.5 42.5 42.2 41.9 40.9
Phase One: collisions snapped to the nearest points on the link-node system
Road names and district matched (%) 46.1 46.3 45.7 48.5 46.0 49.0
Cum. correct (%) 90.6 89.8 88.2 90.7 87.9 89.9
Phase Two: incorrect district board information identified and amended
Road names and district matched (%) 4.5 4.2 4.5 3.9 4.3 3.5
Road names remained incorrect (%) 1.0 0.8 1.1 0.8 0.5 1.0
Cum. correct (%) 95.1 94.0 92.7 94.6 92.2 93.4
Phase Three: unmatched collisions snapped to the next nearest points on the link-node system
Road names and district matched (%) 3.0 3.7 5.3 3.5 5.9 4.2
Cum. correct (%) 98.1 97.7 98.0 98.1 98.1 97.6
Phase Four: incorrect district board information identified and amended
Road names and district matched (%) 0.1 0.1 0.1 0.1 0.1 0.1
Cum. correct (%) 98.2 97.8 98.1 98.2 98.2 97.7
Phase Five: unmatched collisions snapped to the “Identified” road
Road names and district matched (%) 0.7 0.9 0.8 0.7 0.6 1.1
District further amended (%) 0.1 0.2 0.2 0.1 0.1 0.2
Cum. correct (%) 99.0 98.9 99.1 99.0 98.9 99.0
Phase Six: missing road names and typo-errors identified and amended (manually corrected)
Missing road names (%) 0.6 0.6 0.5 0.6 0.6 0.6
Wrong spellings or other typo-errors (%) 0.4 0.5 0.4 0.4 0.5 0.4
Cum. correct (%) 100 100 100 100 100 100
Nature of Spatial Data, Accuracy, and Validation 145
the “identified” road. At step five, almost 98.9%–99.0% of the collision records had been geovalidated using the GIS system with computer programs.
Then, the last step of the system checks for problems related to missing road names in either the collision or the road network database. All these and remaining records are then checked manually for wrong spellings and other typographical errors. With about 15,000 collisions per year, it means that only about 150 collision records (1%) had to be checked manually. With the GIS-validation procedures, the task of spatial data validation has been not only improved but also simplified substantially. In par- ticular, a distinction of junctions and road segments is a big breakthrough in assigning collisions correctly. Moreover, the development of a GIS algorithm to geovalidate automatically means an enormous reduction in efforts to improve spatial accuracy.
During the study period, the majority of the collision records (14,900 or 99%) had been checked and validated by the computerized GIS-based validation system.