Effect of climate and geography on worldwide fine resolution economic activity

greater than the set ‘gradient’ threshold (see Table 1) and is expressed as the percentage of times this threshold is exceeded. This variable reduces the nMAE to ca. 0.23. An increase in air pressure, as expressed by this variable, is normally associated with drier and generally fairer weather. As with MSLP SD S, we see here again that the ‘traditional’ mean temperature and precipitation, while indirectly influencing air pressure do not rank amongst the most impor- tant variables in our analysis. Instead, variations, rather than mean values, appears to be emerging as most critical factors for explaining the geographical distribution of GCP-PC. The fifth variable is the median of solar radiation which reduces nMAE by a further 0.02. Although this variable is less prominent than the variations in air pressure in our results, it is amongst the meteorological factors which has been found to affect mood or behaviour [26]. As with the third variable, the sixth (distance to lakes), the seventh (distance to a river) and the ninth (distance to ocean) most important variables are of geographical nature, as opposed to (purely) meteorological. Instead, the eight and tenth variables, in the top ten list, are again of meteorological nature. As noted above, the marginal reduction of the nMAE beyond the sixth predictor is minimal–nMAE drops by 0.02 from the sixth to the tenth predictor. Note also that air pressure enters the top ten list in three different ways: month-to-month variations, positive

one-step ‘gradient’ and its mean. This is a clear indication that air pressure is a critical variable for explaining the GCP-PC. The maps with the geographical reconstruction of GCP-PC are shown in Fig 4. The top left GCP-PC map only uses latitude (the top variable) as its predictor. This is apparent from the zonal stripes, and with GCP-PC generally increasing with the absolute value of latitude. It is also evident that using latitude by itself, it is not possible to capture regional GCP-PC varia- tions. These are introduced when the second predictor (MSLP SD S) is also considered. Now the regional fit is considerably improved, to the extent that areas in the Arabian peninsula, but Fig 4. Maps of GCP-PC prediction. GCP-PC prediction (log10 of k USD/person) for the global case using the top predictors (see Table 4). Top left: top predictor. Top right: top two predictors. Bottom left: top three predictors. Bottom right: top six predictors. Compare with the actual GCP-PC in Fig 1. https://doi.org/10.1371/journal.pone.0229243.g004 PLOS ONE Effect of geography on fine resolution economy PLOS ONE | https://doi.org/10.1371/journal.pone.0229243 March 2, 2020 10 / 20