The urban sprawl dynamics: does a neural network understand the spatial logic better than a cellular automata?



high highways
main roads
/V other roads
∕ railroads

M -100 --80

M -80 - -60

M -60 - -40

M -40 - -20

M -20 - 0

M 0 -10

M 10 -20

Figure 7 - the differences between observed and calculated values, in blue the underestimations, in
red the overestimations

8. The prediction capabilities of the SANN

Once the learning and testing phase has been concluded, the averaged weight matrix of
the SANNs is processed with the Data set of cells “potentially” in urbanisation in the
next time lag (1994-2008). The prediction concerns “green” cells with urbanized
neighbourhood at 1994.

Figure 8 shows the estimated surfaces for each land use. As expected the trend is linear,
given the availability of only two temporal thresholds.

The resulted pattern shows a probable scenario (fig. 9 - Residence prediction) where
prevailing urbanisation process takes place at the boundaries of the cities and villages

14



More intriguing information

1. The name is absent
2. The name is absent
3. Empirical Calibration of a Least-Cost Conservation Reserve Program
4. The name is absent
5. Stable Distributions
6. Modelling the Effects of Public Support to Small Firms in the UK - Paradise Gained?
7. The name is absent
8. Higher education funding reforms in England: the distributional effects and the shifting balance of costs
9. The name is absent
10. Workforce or Workfare?