Data analysis
The main improvements are:
1. ANISOTROPIC KERNELS The earth's latitudes remain constant everywhere, ie. 1 degree lat is 111km at the equator and in Canada. However, the earth's longitudes vary, ie. ot the equator 1 degree long is 111km but drop to zero at the poles! Therefore each kernel (created by the geographical smoothing out of probabilities from each dig site) has a unique eliptical shape according to its latitude.
2. INCREASED RESOLUTION IN TIME Time slices are parameterised to be able to contain the sum of any number of years. In the examples below each frame consists of summed probabilities of 8 years, giving excellent detail of the rate of local density change.
3. ADJUSTING BY DIG-SITE DENSITY The basis of this study is that 'the number of site phases in a given time period can be expected to relate monotonically to population size' (ref 1). Unfortunately, dig sites are not evenly distributed throughout the British Isles. Therefore the localised density of dig sites will heavily bias the inferred density of local population size. By first calculating the kernel density of digsites through GB, the probability density can easily be adjusted by the digsite density. Indeed, it can be fully adjusted, fully un-adjusted, or any anything inbetween.
4. SYSTEMATIC BANDWIDTH SELECTION Bandwidth is automatically selected from the data by cross-validating using the MISE (mean integrated square error) method.
Various 2D images
un-adjusted by digsite density |
example 1 |
adjusted by digsite density |
example 2 |
Various 3D images
adjusted by digsite density |
example 3 |
adjusted by digsite density |
example 4 |
adjusted by digsite density |
example 5 |
adjusted by digsite density |
example 6 |
Old stuff
example 12 |
|
example 14 |
example 15 |
ref 1: Radiocarbon evidence indicates that migrants introduced farming to Britain. M Collard, K Edinborough, S Shennan, M G Thomas. 2009.