Research demonstrates that urban crime tends to concentrate in certain areas of cities. But, are crimes in these areas spatio-temporarily related or crimes taking place in an area occur independently of what happens in others? Provided that a crime occurs in a certain area, what is the probability that the next crime occurs in other neighboring area? In this study, we try to answer these questions by means of complex networks.
We divide the city of Valencia, Spain, in 599 non-overlapping sectors, which we associate to the nodes of our network. We start with a complete network where all links have a weight equal to zero. Once a crime occurs in a sector, say sector A, we consider all the next crimes that take place in all sectors that are at a certain distance of sector A and occur within a certain interval of time after the event. We increase the weight of the link between sector A and and another sector, say sector B, provided that the next crime occurs in sector B, the distance between sectors A y B is less than a given R, and the time between both events is less than a certain T. Let us call the network thus constructed the real-events-network. The link weights of this network will shed light upon the second question we posed above.
The real-events-network is then compared with a distribution of networks each one constructed following the same, above-mentioned procedure but after randomizing the events in time. Each link of the real-events-network whose weight is smaller or equal than the maximum weight found for it in the network distribution is removed. After this step, we end up by having a sparse network whose links are directly related with spatio-temporal correlations among the sectors where crimes take place.
Despite of relatively long history of point process theory, see, amongst others, Diggle (2013) and Baddeley et al. (2015), few attention has been paid to the analysis of spatial point patterns when the features of interest are curves, i.e.
realisations of some underlying continuous mechanism occurring over point locations. Examples of such spatial configurations include forest patterns where for each tree we have a growth function, economic and population features related to town locations, and spatial locations and curves defined in terms of LISA functions, which define local characteristics of the underlying point pattern. Despite the increased availability of such complex planar mark point process scenarios, there is still a lack of methodology for the case where each point is augmented by multiple function-valued marks. Addressing these limitations, extension of second order characteristics for non-scalar point attributes, as already defined by Comas et al. (2011), Comas et al. (2013) and Ghorbani et al. (2011), are extended to more complex function-valued mark structures. In particular in this work, extensions of the classical mark variogram function, see Cressie (1993), and the mark correlation function, see Isham (1985) and Stoyan and Stoyan (1994), to cross-function version for (multiple) spatial point processes with multivariate function-valued marks are presented. A case study involving the spatial locations of 799 pines (Pinus sylvestris) with some functional-valued tree characteristics are considered to illustrate our new
approaches. Finally, we discuss the results obtained for the case study and propose some future lines of research to extend our approach to more complex non-scalar mark structures.