Effect of climate and geography on worldwide fine resolution economic activity

as well as their variations–particularly mean sea level pressure and dew point temperature–are the main climate predictors that explain economic activity worldwide. Interestingly less than ten variables, and usually six-seven variables explain around 80% of the variance in GCP-PC. For completeness, the middle and bottom terciles of GCP-PC have also been modelled. The simulations and predictions for these two cases provide further evidence that most of the economic activity, as represented by the GCP-PC, can be explained through a (limited number of) geography and climate predictors, even if the accuracy of the results considerably decreases for individual ter- ciles in terms of nMAE compared to the global case. Further, and crucially, the results for the three terciles, as well as for the whole distribution, indicate a close agreement amongst each other in terms of the most important explanatory geography and climate predictors, with six variables, out of ten, in common when all four cases together are considered. This also demonstrates the robust- ness of the non-linear machine learning modelling framework adopted here. While the role of institutions (and natural resources) has been marginally considered in this study, the fact that the global GCP-PC can be reproduced with a high level of accuracy by using a limited and recurring set of geography and meteorology variables, indicates that these are critical factors in explaining fine location-specific economic activities. Aside from the stan-

dard latitude indicator, it remains to be established exactly what the physical links between the identified most important explanatory variables, such as variations in air pressure, and eco- nomic productivity are: these links could be the object of a future study. Such a study would draw on the growing bio-meteorological, physiological and psychological literature that, as presented in this paper, relates meteorological variables and its variations, including of air pressure and humidity, to human mood and behaviour, and which in turn could affect eco- nomic productivity. Our results may have other important implications such as the fact that the relationship between climate and economic activity in the recent past could provide an indication of what the climate conditions were in the distant past in relation to known economically active regions of the world (e.g. the once prosperous Mesopotamia). Conversely, knowing how the climate is projected to vary in the second half of this (XXI) century can give an indication of the possible future economic activity in various parts of the world. Another application could be the consideration of relevant geography and climate conditions to informing where new economic activities could be established to enhance favourable economic outcomes. Supporting information S1 Fig. Map of gross cell product. GCP in log10(k USD); this is also referred to as Gross Cell Product. Note the different scale than the one used for GCP-PC (e.g. in Fig 1). Source G-Econ [18]. (PDF)

S2 Fig. Maps of GCP-PC prediction for top tercile. As in Fig 4 but for the top tercile of the GCP-PC distribution. Compare with the actual GCP-PC in Fig 1. (PDF) S3 Fig. Maps of GCP-PC prediction departures for top tercile. As in Fig 5 but for the top ter- cile of the GCP-PC distribution. (PDF) S4 Fig. Maps of GCP-PC prediction for middle tercile. As in Fig 4 but for the middle tercile of the GCP-PC distribution. Compare with the actual GCP-PC in Fig 1. (PDF) PLOS ONE Effect of geography on fine resolution economy PLOS ONE | https://doi.org/10.1371/journal.pone.0229243 March 2, 2020 17 / 20