Effect of climate and geography on worldwide fine resolution economic activity

Not only is this meteorological variable more important than other geography variables, criti- cally it is also not one that is normally considered in the literature (typically these are mean temperature or precipitation). And while air pressure technically encapsulates information about temperature, it is influenced by several other meteorological variables, like humidity. Importantly, this is not the standard average variable but a measure of variability, in this case an indication of seasonal (or intra-annual) variations. Also, while it would be inaccurate to identify MSLP SD S directly with specific meteorological phenomena such as the passage of a storm, this statistical measure contains the signature of storms. Therefore, this result seems to corroborate the more geographically and temporally limited finding of [2]. Additional evi- dence is discussed in section ‘Patterns of top climate variables and potential physical link with GCP-PC’. The third most important variable in this ‘all grid points’ case is a geography variable, the distance from a major river. This variable reduces the nMAE to 0.28. The implications of the proximity of a major river are evident from a transport, recreational, etc. view point, but it is also worth mentioning that a link exists between rivers and environmental conditions such as water availability or soil conditions, therefore with clear connections to meteorological vari- ables such as precipitation, temperature and humidity. The fourth variable is another meteoro-

logical variable, namely the step-to-step variations of air pressure at 6-hour intervals that are Fig 3. Summary statistics of GCP-PC prediction. Statistics of log10 GCP-PC prediction using 5-fold cross validation, adding one predictor at the time (forward step-wise approach) in the order given and up to 10 predictors. For instance, in the top left panel ‘MSLP SD S’ means that ‘MSLP SD S’ and ‘Latitude’ are used together (cf. also Table 4). Top left: all grid points. Top right: top tercile of GCP-PC. Bottom left: middle tercile of GCP-PC. Bottom right: bottom tercile of GCP-PC. Metrics used are: linear correlation (red squares) and normalised Mean Absolute Error (nMAE, black triangles). nMAE is equal to 0.56 for all grid points, 0.17 for the top and mid terciles, 0.26 for the bottom tercile. https://doi.org/10.1371/journal.pone.0229243.g003 PLOS ONE Effect of geography on fine resolution economy PLOS ONE | https://doi.org/10.1371/journal.pone.0229243 March 2, 2020 9 / 20