Effect of climate and geography on worldwide fine resolution economic activity

due to collinearity of variables (or predictors) is considerably reduced, or even eliminated [24]. The justification for using a non-linear model stems from the complex relationships between any one variable and GCP PC. These have been assessed through scatter plots, also reflected in the low correlation coefficients (see later Table 4) as well as, critically, by the low performance of the multi-model approach (see Table 3). We use around 120 geography and climate predictors (see Tables 1 and 2). By construction many of these predictors are highly correlated. However, since the RF model can deal very well Fig 1. Map of log Gross Cell Product per capita. GCP per capita in log10 (k USD/person). Note that the colour palette has been divided into three sets of three colours each, which represent the three terciles, whose thresholds are 3.41 and 4.05, respectively (cf. Fig 2). Source G-Econ [21]. https://doi.org/10.1371/journal.pone.0229243.g001 Fig 2. PDF of Gross Cell Product per capita. Probability Density Function (PDF) for log10 of GCP-PC. The black vertical line shows the mean of the distribution which is 3.73 (equivalent to about $5350/person). The PDF displays an interesting tri-modal behaviour (cf. also Fig 1): the first mode (between 3.05 and 3.55) corresponds to most of East and South Asia, Central-Northern Brazil, and parts of Western Africa; the second mode (between 3.55 and 4.1) covers Eastern Europe, Northern and Southern Africa, most of Central America and the rest of South America; the third

mode (between 4.1 and 4.65) covers North America, Western Europe, parts of Russia, the Arabian peninsula, Japan and coastal Australia. The darker grey denotes the upper tercile. https://doi.org/10.1371/journal.pone.0229243.g002 PLOS ONE Effect of geography on fine resolution economy PLOS ONE | https://doi.org/10.1371/journal.pone.0229243 March 2, 2020 6 / 20