Effect of climate and geography on worldwide fine resolution economic activity

also India, Central Africa and several other parts of the globe, now display a higher GCP-PC than other locations at the same latitude, and captured by the latitude predictor alone. This improvement is more clearly shown in Fig 5 (top left). Here the change in prediction from one step to the next is expressed as the difference between the individual absolute departures from the actual GCP-PC values for prediction ‘n-1’ minus prediction ‘n’ and calculated as follows: D GCP PC ¼ j Prediction ðn 1Þ GCP PC j j Prediction ðnÞ GCP PC j ð2Þ A positive (negative) value means that the additional predictor at step ‘n’ is improving (deteriorating) the prediction obtained at the previous step ‘n-1’. Although the introduction of the second predictor generally improves the fit, there are also areas where the fit deteriorates, such as in Eastern Europe, North West Africa, Southern Africa, South Western USA, and North Eastern Australia. Some of these, particularly the latter one, are rectified when the third predictor (distance to a major river) is used (bottom left in Fig 4 and top right in Fig 5). Others, specifically Eastern Europe, North West Africa and South Western USA, then improve with the fourth predictor (MSLP positive step change) as seen in the bottom left panel of Fig 5. The bottom right panel in Fig 4 shows the geographical fit with the top six predictors, after which the error decreases only marginally (cf. Fig 3). This is con- firmed by the bottom right panel of Fig 5 which shows that the change in GCP-PC prediction

when the top seven variables are used (compared to the top six) is close to zero in most areas of the globe. While some noticeable differences between the predicted (using the top 10 predictors) and the measured GCP-PC can be seen (Fig 6 top left), these appear to be due, for the most part, to non-systematic errors. Possible exceptions, with consistent discrepancies either over a single Fig 5. Maps of GCP-PC prediction departures. Changes in prediction expressed as the difference between the individual absolute departures from the actual GCP-PC values for step ‘n-1’ minus step ‘n’ for the global case. A positive (negative) value means that the additional predictor is improving (deteriorating) the prediction. Top left: Change from step 1 to 2. Top right: Change from step 2 to 3. Bottom left: Change from step 3 to 4. Bottom right: Change from step 6 to 7. https://doi.org/10.1371/journal.pone.0229243.g005 PLOS ONE Effect of geography on fine resolution economy PLOS ONE | https://doi.org/10.1371/journal.pone.0229243 March 2, 2020 11 / 20