Effect of climate and geography on worldwide fine resolution economic activity

Dew Point bottom Q (0.36), Solar Radiation bottom Q (0.36) and latitude (0.35). The first two and latitude occur in the all-grid-point case, whereas the other two are for the middle tercile. It may appear odd that despite Solar Radiation median and MSLP positive step-one change have a higher linear correlation with GCP-PC than latitude in the all-grid-point case, it is the latter that ranks first in importance. This behaviour may be explained through the following assessment. For illustrative purposes, we consider the quadratic fit of the top ten predictors for the all- grid-point case. While generally there is a wide scatter between each predictor and GCP-PC, it also apparent that the relationship between individual predictors and GCP-PC is not a linear one (Fig 7). Thus, whether the most appropriate line fit is quadratic or a more complex one, these plots give an indication as of why a non linear model such as RF is superior to a linear model. At the same time, it is important not to over-interpret the results of the quadratic fit. If we take for instance the third variable, distance from major rivers (second top row, left), it would appear that at distances larger than around 4000 km, namely the minimum of the qua- dratic curve, GCP-PC starts to increase again. This is essentially due to the lack of major rivers, as defined by [21], in countries such as Australia and New Zealand (see S8 Fig). On the other hand, the quadratic fit can provide some useful indications about the values of the variables

that are linked to above or below GCP-PC average. For instance, in the case of latitude above average GCP-PC appears north of around 40˚N and south of 30˚S (see where red line crosses the grey line in Fig 7); in the case of MSLP SD S, higher than average GCP-PC appears for val- ues between about 2 and 7 hPa; and for MSLP positive one-step change, above average GCP-PC appears for values between about 2.5 and 17.5%. This assessment is complemented by the regional geographical patterns of the top ten predictors presented in the supplementary information (S8 Fig). Although the majority of the literature linking economic productivity-related factors such as physiology or labour allocation with weather and climate take only temperature, and some- times precipitation, into account [27, 28, 29, 30], there is increasing evidence that meteorologi- cal variations affect human mood. It is long been known that mood, which can be categorised as concentration, cooperation, anxiety, depression, sleepiness, and other behaviours, can have significant impacts on human activities, and ultimately economic productivity [31]. Perhaps the most comprehensive reference for the effect of meteorological (variations) on mood is [26]. In this book, we read for instance that ‘it is a more complex mix of weather vari- ables including pressure drop, humidity and/or temperature that causes the greatest stress on the body and most likely leads to increased pain’. Or that ‘infrasound, also caused by fluctuations in

atmospheric pressure, can easily penetrate buildings and affect their inhabitants’ [26]. Physiol- ogy explanations of the effect of weather on human body date back to the 1930s with [32], where it is stated that when a weather ‘front approaches and the atmospheric pressure begins to fall, the body responds by contracting blood vessels and reducing the amount of oxygen to the heart, brain, kidneys and other major organs. This causes blood pressure to rise, resulting in an overall ‘stimulation’ of the body’. More specifically, [33] finds that pressure change, even more than pressure level, appears to be important in determining mood/behaviour. Similar findings are reported in [34, 35, 36]. Interestingly, [35] also found that effect of meteorological factors on mood is nearly the same whether people spend almost all of their time indoors (i.e., less than 30–45 min outside) or outdoors. Moreover, [36] found a correlation between the fre- quency of migraine headaches and the day-to-day difference in barometric pressure. In the context of our results, and referring back to the identified ranges in the variables in Fig 7 and S8 Fig, their values appear to imply that higher economic productivity is linked with variations in meteorological variables that are moderate: too low values may not provide suffi- cient physiological and/or psychological stimulation [26], whereas too strong values, likely PLOS ONE Effect of geography on fine resolution economy PLOS ONE | https://doi.org/10.1371/journal.pone.0229243 March 2, 2020 14 / 20