Geocoding Accuracy - How much accuracy do you need?
Tom Turchioe
Chief Architect, SAP Global Technology Channel, TRC | Innovation, GIS and GeoAI Strategy
In my last blog, I described the basics of geocoding (also called geoenabling data). In this Holiday blog, I take a look at geocoding accuracy. The more accuracy desired, the more it costs. Your business requirements should drive how much accuracy you need. There are at least 2 major methods: rooftop accuracy and street-level accuracy. In the image above, the street address is 300 Pasteur Drive, Stanford, CA. Enter that address into Google Maps, go to Street View and you'll see this:
The address on the glass above the door is, indeed, 300 Pasteur Drive, but Google reports that the address is 318 Pasteur Drive. Why is that? It's because Google Maps uses street-level accuracy. It's typically obtained using the U.S. Census TIGER dataset which uses address range interpolation that works like this: if a given street block has addresses ranging from 300 to 400 and you search for 370, the returned lat/long will be about 70% along the block starting from 300. Navigating in Street View until I find 300 Pasteur Drive, I will see this:
Because Pasteur Drive is short, using address ranges means the accuracy is good enough that you will find where you are going. On a longer block, with unevenly spaced buildings, it might be harder to find the desired address but the pizza delivery person will likely find your home. Here is an example of looking for 407 Highway 28 (a real home I lived in) somewhere in the Sierras. The following image shows the Street View for the closest address:
It turns out the actual home is 400 ft. to the south on Highway 28. When the address is typed into the Google Maps search box, the map view shows the pin about 400 ft. south of the actual structure. The houses along this stretch of Highway 28 are very unevenly spaced and not consistently numbered which accounts for the greater inaccuracy. There are other limitations of address range interpolation including the address not being platted: 407 1/2 Highway 28 may not exist, but the geocoder may still provide a lat/long anyway. The point here is to understand the limitations of address range interpolation and factor that into the level of accuracy needed, and determine if it suffices for your needs.
The second method is rooftop accuracy. The lat/long of the rooftop can be determined from a satellite image at a high enough resolution or from aerial photographs taken from airplanes or drones. Feature extraction can be used to identify rooftops and determine the lat/long. This is a very accurate method, but it's still possible that the wrong roof (of an outbuilding instead of the main structure) may be tagged resulting in an incorrect lat/long. Here is an example:
If the lat/long of the outbuilding at the top of that plot is provided, that's clearly not what was desired.
A quick word on data quality - inputting different variants of a specific address may lead to different results. If the geocoder can't determine correct lat/long based on the provided address, the pin on the map may be placed at the centroid (center) for the specified town or city or the pin may be placed at a street of a similar name in a completely different town, city, state or country.
More and more organizations are geoenabling their enterprise data. Its impact on improving business processes is beyond dispute. Just be aware of the different methods and their accuracy with respect to your needs and always pay attention to data quality.
Have a great Holiday Season. I'll be back in 2017!
Managing Director, Energy/Utilities EAM, Field Service & Mobility Systems Integration
8 年Excellent comments by all on this topic, including the concept of MDG. LAM could augment GIS with the use of offsets and landmarks for critical point assets which could serve as the "origin" for a group of assets in proximity.
Technical Director, Operational Technology Architect @ Amtrak
8 年Mark, Whatever level of accuracy is prescribed, SAP Master Data Governance (MDG) for Linear Assets can ensure geo-spatial and linear equipment attribution is aligned. SAP MDG for EAM Version 730 has just passed quality and includes ability to master Maintenance Plans, Task Lists with linear - ideal for advanced digital and maintenance strategies. Merry Christmas.
Chief Architect, SAP Global Technology Channel, TRC | Innovation, GIS and GeoAI Strategy
8 年I agree with Mike: is close enough good enough? You mention they are doing analysis and modeling. If it is some kind of roll up or aggregation, then high accuracy may not be needed. For instance: how many transformers and what is the average number of repairs in the last year in a given town or what's the cost of repairs spent on a given type of equipment in southeast Houston during storms? Then a few miles error on some small percentage of assets may be acceptable.
Helping Senior Leaders take advantage of location
8 年Dan - my $0.02 on this goes back to the value of the precision to the business. If close enough is good enough then getting the enterprise to pay for precision that is not valued is not a good use of their capital. However, a good ROI, based on something that your client values is likely the right path. Lately we have been having to describe the value of Geo locating assets for maintenance activities (think utility poles). Some utilities are ok with "get our drivers close enough to see them" and others needed something more like "I need to know what side of street and beside which house".
IT Strategist and Consultant
8 年I'm curious about your thoughts on business requirements and how they might drive the need to manually improve low accuracy placement. we're geocoding thousands of sites for a client and dealing with that exact issue right now. I'm making the argument that for the precision of their models and analysis, an address being even several miles from it's true location, in some cases, is acceptable vs. the labor LOE to correct.