To trade or not to trade: Link prediction in the virtual water network
Given the dynamical behaviour of the virtual water trade network, where links are created and dismissed every year, predicting the link activation/deactivation is an open research question. We propose a novel methodology for link prediction applied to the network of virtual water trade. Starting from the assumption of having links between any two countries, we estimate the associated virtual water flows by means of a gravity-law model using country and link characteristics as drivers. We consider the links with estimated flows higher than 1000 m3/year as active links, while the others as non-active links. With this approach, we are able to correctly model 84% of the existing links and 93% of the non-existing links in year 2011. Results indicate that, over the period from 1986 to 2011, population, geographical distances between countries, and agricultural efficiency (through fertilizers use) are the major factors driving the link activation and deactivation. As opposed to other (network-based) models for link prediction, the proposed method is able to reconstruct the network architecture without any prior knowledge of the network topology, using only the nodes and links attributes; it thus represents a general method that can be applied to other networks such as food or value trade networks.
Through the international trade of agricultural goods, water resources that are physically used in the country of production are ‘virtually’ transferred to the country of consumption. The volume of water virtually transferred is known as ‘virtual water’. The virtual water perspective shows up the interdependency existing across water resources worldwide.