Positional accuracy and integration of geographic data
In Great Britain most geodata that provides the reference and geographic context for more targeted user datasets created by individual organisations is issued by Ordnance Survey®. These user datasets may include Sites of Specifi c Scientifi c Interest or Basic Land and Property Units. Based on this reference data, users often integrate other datasets, such as statistical tables or their own geo-datasets, to support analysis and decision making. Others may collect geo-datasets by using Global Positioning System (GPS) equipment: the location of street furniture, for example. In all cases it is vital that data from different sources fi ts together spatially to enable the joint use and analysis.
There are two main drivers to change geodata. Firstly, there is the requirement to incorporate real-world change (RWC) such as new houses, roads or demolitions. By its nature, this results in additions or deletions to existing holdings. The second driver is when changes to technology make wholesale improvements to existing datasets a benefi t or necessity.
Positional Accuracy Improvement
While changes in the real world usually have a low density and are taken into account by frequent updates of the reference data, other changes, such as the ones triggered by the Positional Accuracy Improvement (PAI), illustrate the necessity to manage dataset against reference changes in a far more organised way.PAI and update issues are not specific to Great Britain. EuroSDR’s international PAI workshops (www. dit.ie/eurosdr) have proved that the same issues are present in most countries with an advanced availability and use of GI. The experience in other European countries has shown that managing change to achieve interoperable data is a natural part of the evolution of geodata and its use within GIS. Before looking into the British scenario in a little more detail, the following examples illustrate the width of PAI-type issues:
USA: MAF/TIGER Accuracy Improvement
In the United States the most prominent PAI programme is undertaken by the US Census Bureau as part of the MAF/TIGER Accuracy Improvement Project (http://www.census.gov/geo/mod/ maftiger.html). The programme aims at improving the accuracy of the TIGER (Topologically Integrated Geographic Encoding and Referencing System) database to 3.8 metres root mean square error (RMSE). In contrast to the current data, that has been reported to differ up to 150 metres from its (true) Differential-GPS position, this will benefit users of the data such as Local Government to allow them to insert census geography easily, use GPS on handheld computer and remove the need for paper maps, and to enable field staff to relocate a structure saving time and cost per case. The programme has an investment volume in excess of $200 million.
Germany: Integration of utility asset records
In a more localised scenario DEW, a German utility company based in Dortmund, was confronted with the problem of inheriting two different cadastral reference maps for the electricity and the gas/water assets after a merger. DEW successfully finished shifting more than one million GIS objects of the gas/water information layer to the reference of the new reference map using a sophisticated software tool. This example shows how water and gas pipeline data is created and displayed against cadastre (or alternatively topographic) data. While a lot of network assets are surveyed against real word objects such as house corners, most of the data that is used in the government sector—for example planning applications—was digitised against the reference map. As most Geographic Information Systems (GIS) store data in independent layers with geometry information as coordinate strings, no relationship information (or the original measurements as in the utility example) is retained after the dataset is created. If the reference data is positionally improved, the relationship between the reference data and the overlay user datasets is destroyed.
Ordnance Survey’s PAI programme
It has been apparent since GPS was first used as a surveying tool that highly accurate GPS coordinates cannot always be seamlessly integrated into the map data ofthe British National Geographic Database. This is due to a fundamental difference in Absolute Positional
While differential GPS methods allow Absolute Positional Accuracies of a decimetre or better, features in largescale Ordnance Survey map data have an Absolute Positional Accuracy of between 2.8 m Root Mean Square Error (RMSE) in rural areas and 0.4 m RMSE in urban areas. This indicates the accuracy of the absolute position of a coordinate in the context of the British National Grid coordinate system. In contrast to this, the Relative Positional Accuracy between features – two houses, for example – has always been signifi cantly better.
Following earlier debates that go back to about the 1970s, Ordnance Survey started to plan a national programme to improve the Absolute Positional Accuracy of its rural large-scale map base at 1:2500 scale in the late 1990s. It applies to 152,000 km2 (or about two thirds of the area of Great Britain) and excludes the major urban areas, which were already resurveyed to a higher standard from 1947 onwards, as well as mountain and moorland regions, where a high Positional Accuracy is not necessary. The first block of improved data was released in November 2001 and the programme is scheduled to be completed by March 2006. As of July 2005 about 80% of the data has been issued. The aims are to future proof the large-scale topographic database for the addition of new building development and other change as well as providing a better relationship between Ordnance Survey’s rural map data and users’ own GPS-positioned data. The Absolute Positional Accuracy of the data after the improvement will be 1.1 m RMSE in rural areas and 0.4 m RMSE in selected rural towns. Following an analysis of the original and improved data it was found that the shifts can not be modelled in a mathematical way. The majority of shifts are less than 2.5 m in most areas, with only a very limited amount of extreme shifts of up to about 10 m.
Prior to the start of the programme an extensive consultation process was conducted. Over the last four years Ordnance Survey has continued this communication with over 10 seminars, small workshops with select user groups, one-toone dialogue and close liaison with system suppliers and solution providers. This work, in conjunction with engagement with representative bodies in government, has led to the development of a number of guidance documents and case studies, which are published on Ordnance Survey’s dedicated PAI website www.ordnancesurvey.co.uk/PAI.
PAI for data users
Data users in Great Britain have learned that PAI may have a signifi cant impact on their use of digital geographic data. In particular, automated searches, such as land charge searches for conveyancing, may produce different results if an initial search has been done on a pre-PAI reference dataset and a subsequent one, a few years later, utilises a post-PAI reference map. In the example of a contaminated land search there is the possibility for litigation if an authority knowingly uses datasets of different accuracy that either wrongly reduce the value of a property or miss vital information. Once a user dataset is shifted to its post-PAI position, it can safely be used in conjunction with a post-PAI Ordnance Survey reference map, but not necessarily with other user datasets that haven’t been shifted.
For searches that incorporate user data from different data providers, it is important to verify the PAI status of these external datasets before use. Therefore standardized metadata about the Positional Accuracy status of datasets or, maybe, even individual features is desirable.
Since the data is positionally improved and updated for RWC at the same time, all subsequent map updates will be based on the improved data. With about 5,000 km2 of PAI improved data to be released every month until March 2006, users are receiving more and more PAI data. The majority of Ordnance Survey’s largescale data users in rural areas are currently either working on strategies to use positionally improved reference data or actively using PAI data already.
An example: British Waterways
British Waterways is one of the organisations that has shifted all areas released to date. Martin Rivas, GIS specialist at British Waterways sums up his experience of implementing PAI in an area of 700km2: “The key to successful PAI implementation is to employ simple tools and a simple process as well as being able to rely on a robust system infrastructure. In addition to necessary data cleanup and quality assurance, it took on average two hours to shift 45 datasets in an area of about 20 km2.
British Waterways put an emphasis on the relationship between their various datasets and used topology rules (rules such as ‘polygons of waterway features must always be polygons of the descriptive group inland water in Ordnance Survey’s OS MasterMap® digital map product’, http://www.ordnancesurvey.co.uk/ oswebsite/products/osmastermap/) to validate them against each other.” Martin Rivas also states that “Using topology rules means that we are certain to correctly maintain the exact internal relationships”.
Positional Accuracy Improvement happens on two levels: reference and user data are adjusted to generate datasets that can be used in conjunction with each other and are compliant with GPS measurements as well. Saying that, there is still a difference between the Absolute Positional Accuracy of a few centimetres that can be derived from GPS measurements and the improved accuracy of (1.1 m or 0.4 m RMSE) of the large-scale Ordnance Survey data. At this point in time and for the foreseeable future this accuracy is sufficient for the majority of uses and can be economically maintained by existing and proven technology.
The experience with PAI so far has shown that the creation of improved reference maps as well as the adjustment of user data requires an investment into data from both reference data providers as well as the users of this data. On the user side, it has become acknowledged that PAI is part of a wider data management strategy that is not sufficiently addressed by a number of data users; in fact, very few users manage RWC. As long as the data is kept in a well-guarded environment, like a small GIS team, this hasn’t proved to be a big problem, but with the move to corporate information systems 1,000s of Intranet users and many potential Internet users will be accessing uncontrolled overlays of various datasets of different accuracy. This opens up an huge potential for misinterpretation and means that it will be important to manage the integrity of datasets against each other.
The management of both RWC and PAI illustrates a simple fact: geographic data used as a reference to provide spatial context needs to continuously refl ect the changes in the real world but, on the other hand, needs to be stable enough to ensure that the reference is not lost over time. In OS MasterMap this is supported by the existence of Topographic Identifi ers (TOIDs) and feature life-cycle rules to enable users to manage these changes in respect of their own data.
Positional Accuracy is an issue of Geometric Interoperability that is becoming more and more relevant in many countries and is largely triggered by the widespread use of GPS as a very accurate surveying and positioning technique. PAI is often seen as a painful exercise. In practice, supported by British Waterways’ experience, the preparation and planning can be quite complex but the application is fairly easy and straightforward, even for a bigger organisation with thousands of users.
For data users PAI implementation is a necessary investment in maintaining their data holdings. If applied correctly, it will deliver improved data management capabilities and will allow better data integration to empower organisations to make better decisions.