What Is GIS?

In its simplest form, GIS is computer software that links geographic information with descriptive information. Linking information to geographic data is naturalsay 90210, and most people will know that is the ZIP Code for Beverly Hills, Calif., and it represents an affluent segment of the population. (Look, I never watched the show, and even I know what 90210 is, but I dont know why Kelly left Dylan for Brandon.) In the property/casualty business, knowledge about the location of the insured property is of paramount importance. The coverage for a beach-front condo in hurricane country is going to be a lot different than my hunting cabin in bear country.

Think of GIS as a traditional map with layers. Begin with a regular Mercator projection, and start layering transparencies over it. One layer may represent congressional districts, the next layer average annual snowfall, and the next layer the number of violent crimes per capita, etc. You get the picture. We probably all have used GIS software without even thinking about it. Fire up Microsoft Streets and you have access to multiple segments of data layered over the basic road map. You can find the closest airport or Chinese restaurant. The Yahoo Yellow Pages information service is a GIS. Based upon a given location or ZIP Code, it is able to find businesses and their physical relation to that specified location.

So far this all sounds pretty simple. People have been adding information to maps for hundreds of years. The first oceanic charts or maps displayed information beyond land masses, for example, prevailing currents, winds, and impediments such as the Sargasso Sea. GIS is a lot more than just computer software. It is a computer-based system that links disparate types of data (spatial and attributes of that spatial data) in a meaningful way.

Where Am I?

There are two data types in a GISspatial and attribute data. Lets consider spatial data. (For the sake of this brief discussion, we will ignore elevation when discussing spatial data. Obviously, in the real world, we cannot ignore it. Elevation can be treated as an additional dimension to a vector model or as a data attribute.) In general, we are able to pinpoint any location on the surface of the earth using two coordinateslatitude and longitude. Latitude is the angular distance measured in degrees north or south of the equator (0 to 90 degrees north and south). Longitude is the angular distance east or west of the Greenwich Meridian, measured from 0 to 180 degrees east or west. On a standard Mercator projection map, the lines of latitude and longitude are represented as grids. A specific location on the earth is represented by the intersection of two lines (or angles of arc). Translated to a software model, this would be called a vector model. A particular place on a map in a software system is defined as a coordinate pair. The vector model most closely resembles what we may call a traditional cartographic representation. A traditional map represents a series of nodes or vertexes that are always in the same relation to each other (Pittsburgh always may be represented by a vector of X degrees and Y kilometers from Harrisburg). Vector maps are very good at representing accurate spatial relationships on the earth. In fact, they can be exact maps of an actual image. They are not as efficient for storing useful attribute data.

Each vertex represents a unique location, and therefore any data associated with that particular vertex is, in effect, a singularity. What we really want to know is not the crime rate for 33 20 N 44 26 E but the crime rate for an area surrounding that point. In order to extrapolate that data, we must have a database that stores the location of many vertexes and their associated data. We then must apply or develop sophisticated algorithms to manipulate the data associated with a defined region around the node in which we are interested. This sounds overly complicated because it is. We are talking about computer systems here, and all data needs to be reduced to a method whereby a data system effectively can mine that data. I know I can glance at the appropriate map and determine immediately if a given location is in the flood plain for the Susquehanna River. Computer systems are not intuitive. They require a method if they are going to accomplish anything. (Whatever happened to AI? Some of the simplest brain functions resist all attempts at replication by digital machines.)

The World as a Cube Farm

Digitally there is an alternative to the vector model for spatial data. It generally is referred to as raster data. Raster data models incorporate the use of a grid-cell data structure where the geographic data is divided into cells identified by row and column. Divide the entire world into a checkerboard and you have a raster model. A specific location is defined by its location in a particular cell. Pitcairn Island is no longer located at 25 04 S 130 06 W but in cell [XYZ,UWV]. Thus we lose some of the geographic precision available in a vector system. That loss becomes a gain as we digitize the information. The grid-cell model lends itself very readily to computer operations. Two-dimensional arrays are able to be manipulated quickly and easily in any data system. This makes raster models well suited to mathematical analysis. We also have a larger and more manageable sample size for attribute data. It is relatively easy to assign data to a measurable area. For instance, crime patterns within a certain segment of a city readily are available. Continuous and discrete data is well accommodated with a grid structure. For a raster system to operate efficiently, the cell size needs to be considered. Too small a cell and we are stuck with meaningless discreet information; too large a cell size and we have nothing but generalities. There are other drawbacks to a raster model. Input (e.g., real world maps are in a vector format and) must be converted to a raster model. Likewise, mapping data patterns out for visual presentation is more cumbersome than with a vector model.

It readily is apparent no single simplistic spatial data model will fulfill all requirements. Vector systems need to be able to mimic the behavior of raster systems and vice-versa. And no system is worth anything without meaningful attribute data. Insurance companies use GIS systems to visualize, analyze, and distribute risk. We can gain information about our customers purchasing habits and financial behavior and target additional products accordingly. We can analyze a particular region of a city to determine home sales in the last 18 months above a certain threshold value in an effort to target new customers.

Whos Got the Data?

Where does all this data come from? There are private sources of data, but the vast majority of data is available from government agencies such as the Census Bureau, U.S. Geologic Survey, FEMA, and Department of the Interior. Most local governments have a wealth of readily available data on property sales, property taxes, school districts, voting habits, etc. The process of merging attribute data with spatial data is called geolocation. I suspect the most successful users of GIS first will be those who are able to mine the existing data effectively. Following that, organizations will need to create their own custom data and incorporate that into their GIS. Insurance companies probably have more data on human events than any other group of businesses. Incorporating that data into GIS may well spell success or failure in the future.

There is, of course, a practical limit to the amount of data we reasonably are able to process and use. GPS and RFID (radio frequency identification) devices soon will flood us with so much data we will need to be cleverly selective about what we want to record and then what we want to analyze. With on-board GPS receivers and data transmitters, we have the ability to assign risk and charge drivers for insurance on a pay as you go basis. Does that mean we also can change the day premium on a high-rise apartment building because the main water supply for that area is out for the next 48 hours due to a pipeline rupture? The ethical issue here is not my concern, but the information overload issue is. There is a practical limit to what we can process. Geographic information systems store data in relational databases, and while we do have the ability to process terabytes of data, there is a cost-effectiveness ratio we need to consider. We want to fine-tune our business, not micromanage it down to the minutest detail (at least in my humble opinion).

Feeling Paranoid?

A not insignificant issue is responsible use of data. GIS easily could be used to red line certain geographic areas or neighborhoods. There already is a trend in managing risk to limit exposure to certain types of risk. Are we interested in offering coverage only to those segments of our customer base where we find the most profitable return on risk, or do we spread the risk over a large enough pool to smooth out the risk? These are both ethical and business issues. Mississippi has been labeled the fattest state in the U.S. Should Mississippi residents be placed in a different risk category than those from Maine? Should I pay a higher life insurance premium because I work in a state with one of the highest per capita smoking rates? Should I pay a lesser premium because I am a marathon runner (some would say I should pay more)? There is a fine distinction between managing risk and eliminating it. Part of our challenge in using GIS is to make that distinction and then to convey that to our customers. Everyone is a bit wary about big-brotherism, and we need to ensure our data systems do nothing to enhance that fear.