Kartographie 5

symbolisation of areal data, optimal fit between symbol properties and spatial data properties

- data model: 3D, continuous: ratios, densities

- graphic model: 2 D: discrete: planear surface, boundaries unrelated to data, standardised! by e.g. people per km^2, per total population ...

3 D: stepped prism

classed map

pros:

- when phenomenon has distinct breaks or discrete distinctions

- when data distribution shows a particular pattern, such as natural breaks, several peaks, some trend

- also useful because human brain works like that, clear pattern

cons:

- data more aggregated (increased statistical error)

- perceptual limit of number of classes -> 5-7 appropriate

unclassed map:

pros:

- less aggregated, smaller statistical error

- represents better continuous statistical data, graphic model= data model

cons:

- too many individual values are not easy to differentiate (increased human error)

- graphic model not like mind model

- pattern is distribution dependent, thus map comparison is hard

idiographic (greek) = descriptor of uniqueness

- natural breaks: look for gaps in the array of value (in series of ranked observations)

after Jenks: minimize variance in the class, maximize distance between class or:

GVF ( sum of squared deviations between classes) is maximized, squared distance from the class mean is minimized

-quantiles: put equal number of observations (N) in each class (I-N classes)

e.g. round numbers, equal steps

put equal value range (along z-axis) in each class

the error computed can be shown as a statistical surface, a blanket of error map

computed discrepancy between each value(xi) and its associated class mean (x-), similar to the root mean square error

With optimizing class scheme (Natural Breaks Jenks) error should be minimized!

But: error measure is sensitive to number of classes & classing scheme!

-< if necessary: modifyinng classing scheme

visual variables color hue and value applied to choropleth maps