Of the available population density datasets [2], [13], [19], we found the LandScanTM 2007 data [13] (see Supporting Information S1) to be the most suitable for our analysis as it provides information at the most spatially disaggregated level. LandScanTM has a resolution of 30″ (∼1 km at the equator), and the population distribution is based on census data compiled using a multi-layered spatial modelling approach [13]. The main input data are: census information; administrative boundaries; land cover; coastlines; elevation; and imagery [13]. According to the documentation of the dataset [13], the distance to water was not part of the modelling parameters. Therefore, the data are not biased in that sense and can be used for our analysis.

The LandScanTM 2007 dataset does not, however, provide any delineation between urban and rural population. We therefore used two separate datasets to identify the urban, peri-urban, and rural areas (Table 1). According to Potere et al. [23], the MODIS 500 m resolution global urban map [24] has the highest accuracy for mapping urban areas. Therefore, we selected this dataset to identify urban extent. We then used the GRUMP dataset [25] (which was assessed by Potere et al [23] as the least accurate presentation of urban settlement), together with MODIS 500 m data, to identify the peri-urban areas. The peri-urban area is here defined as the area of the GRUMP dataset that is not covered by the MODIS 500 m urban extent area. The area that is covered by neither the MODIS 500 m global urban map nor by the GRUMP data is defined as rural area (Supporting Information S1).

For our analysis we used four classes of freshwater bodies, namely: lakes, large rivers, medium rivers, and small rivers (Table 1; Supporting Information S1). From here on we refer to these classes as water feature groups (WFGs). The spatial data for these water features are based on the VMAP0 (Vector Map Level Zero) dataset [26]. Only perennial water bodies were included in the analysis; wetlands and seasonal rivers were excluded. Of course, in some regions populations do rely on these ephemeral water sources; for example, they have determined the seasonality in farming in the Middle East for millennia [27].

The freshwater bodies were mapped using VMAP0 polygon data, which have a scale of approximately 1∶1,000,000 [26]. From the VMAP0 data, we extracted permanent lakes and large rivers using the Global Lake and Wetland Database (GLWD) [28]. In the latter database, large rivers are derived from the Level 2 data of GLWD (i.e. GLWD-2); this dataset contains the shoreline polygons of permanent open water bodies with a surface area ≥0.1 km². The medium and small rivers were extracted from the VMAP0 data using the World Data Bank II (WDB II) dataset [29]. This datasets has a resolution of 1∶3,000,000; those rivers identified in the WDBII were extracted from the VMAP0 data and classed as medium rivers. The remaining rivers in VMAP0 (i.e. those that were not classed as large or medium rivers) were then classed as small rivers.

The VMAP0 river network is homogenous for most regions. However, for parts of South America and Asia, there are some differences in the level of detail in the mapping of the network. Despite this shortcoming, we believe that VMAP0 is still the most suitable dataset to be used in this kind of analysis. The recent HydroSHEDS (Hydrological data and maps based on Shuttle Elevation Derivatives at multiple Scales) data [30] have higher accuracy than the VMAP0 data, but the HydroSHEDS data do not cover the entire globe (the dataset cover only areas south from Latitude 50°N), and thus the dataset is not suitable for this study.

Due to data availability, the small streams, local surface waters, and temporal water bodies including wetlands were excluded from our analysis, although they are vital sources of water and livelihood in many parts of the world. Groundwater abstraction is also an important source of water in various regions [31], [32], but is not included in this analysis due to poor data availability. Neither does our study take into account the state of a water body in question, although water of poor quality may not be usable at all. Such information should be included in future global analyses if appropriate global data become available. The scale of the data used in the study should also be taken into account when interpreting our results.

Using the dw_land dataset, combined with the population density dataset, we were able to calculate the ‘population distance to water’. This dw_pop corresponds to the median distance of a person to the nearest freshwater body. We calculated dw_pop for all geographical scales presented above, and for each different population and water feature group.

For each FPU we also analysed whether people lived closer to, or further from, freshwater bodies than the average dw_land for that FPU. This was assessed using the ratio of dw_pop over dw_land, referred to hereafter as dwr_pop/land.

We also analysed the average population density and cumulative population as a function of dw_land. This was carried out separately for each population class and for each climate zone. With this analysis we aimed to visualise how population densities change with increasing distances to water, and to illustrate how this differs between population groups and climate zones.

We used several climate variables (precipitation, temperature, and aridity index), together with dw_pop and dw_land, to explore whether population density could be explained by these physical characteristics. Bivariate correlation and multiple regression analysis tools of the SPSS programme (version 19) were used to analyse the correlations between the variables in question.

Finally, we used estimates of water resources availability per capita from Kummu et al. [17] to examine relationships between population distance to water and water scarcity. Data for these two variables per FPU were used to construct a 3×3 matrix with the following thresholds:

Globally, just over half of the population (53%) lives in rural areas, whilst rural areas account for 94% of the total inhabited area (Table 2). On the other hand, whilst about 47% of the world population lives in urban and peri-urban areas (according to our division), these areas account for just 6% of the total inhabited area (being 1.6% of the total land surface area on Earth). On this global scale, the median dw_pop shows very little difference between urban, peri-urban, and rural populations (Table 2); however, there are differences between regions, as will be presented and discussed later.

Moreover, if we examine how population density changes in relation to the dw_land, we see clear differences between the population classes (Figure 2). In total, around half of the world's population lives within 3 km of a freshwater body, whilst 90% lives within 10 km. Globally, average population density gradually falls from over 150 persons/km² in areas closer than 2 km to a freshwater body, to around 50–60 persons/km² in areas at a distance of 25 km from a freshwater body (Figure 2, bar graph). This reduction in population density as dw_land increases appears to be attributable to the situation in rural regions. Figure 2 (line graphs) shows that the population density remains rather stable as the dw_land increases in urban and peri-urban areas, whilst a clear decrease is seen for rural areas. This would seem to suggest that proximity to freshwater bodies is more defining for where people live in rural areas compared to the situation in urban and peri-urban areas.

For the majority of the world population (66%) the closest water feature is a small river, while for only 6.5% of the population it is a large river (Table 3). The population density is highest, however, in inhabited areas where the closest water feature is a large river (Table 3). Based on the results derived from the datasets used, humans inhabit about 38% of the total surface area of the globe (Table 3). For those areas where a river is the closest water feature, humans inhabit over 40% of the area, while for areas where a lake is the closest water feature, only about 21% is inhabited (Table 3). This can be explained by the fact that many of the areas in which a lake is the closest freshwater feature are located in sparsely populated regions in high northern latitudes or in deserts or arctic areas (Supporting Information S1).

The median distance of population to water varies between the WFGs from 2.2 km (large rivers) to 4.6 km (lakes) (Table 3; see also Supporting Information S2). The relatively large distance to lakes can be explained by the same reasoning as the low inhabited ratio (see above). The relatively low population distance to water associated with large rivers can be related to the large population density, which appears to congregate around (inhabited sections of) large rivers.

The decrease in global average population density as dw_land increases is shown again in Figure 3. In this figure, however, the cumulative population living in different climatic zones is also shown, revealing considerable differences between the climatic regions. Whilst on a global scale about 70% of the population lives within 5 km of the closest water feature, this is around 80% for temperate and cold regions. On the other hand, only 55% of the population in arid areas lives within 5 km of the nearest water feature. Hence, in these areas, where water is already (by definition) scarce, the distance to those scarce sources is also relatively large. The median distance to water in arid zones is 4.3 km, compared to 2.8 km in cold and temperate zones.

According to our analyses, people live on average closest to water in Australia and Oceania (median dw_pop 2.3 km), followed by Eastern Europe, Central Asia, Southeast Asia, and Western Europe (2.6 km) (Figure 4). People in Northern Africa (4.3 km) and Middle East (4.8 km) live, on average, the furthest from water (Figure 4).

The clearest difference in median dw_pop between population classes can be seen for North Africa, where the distance to water in rural areas is more than double that in urban and peri-urban areas (Figure 4). In most regions, the dw_pop for urban populations tends to be rather similar to the dw_pop for rural populations (difference less than 0.5 km). Interesting differences are in the Middle East and North Africa, where urban populations live closer to water than rural populations, contrasting with the Americas, where rural populations live closer to water than urban populations.

We also calculated the median dw_pop for each country with more than 100,000 inhabitants. According to the results, people in Suriname live closest to water (median dw_pop was 1.6 km); the median distance is also less than 2.0 km in Kyrgyzstan and Tajikistan. The people of Libya live, on average, the furthest from water (233 km). All the country results are presented in Supporting Information S2.

As described in the methods section, the ratio dwr_pop/land per FPU was used to examine whether people live closer to, or further from, freshwater bodies than the average land distance to water for that FPU. These ratios are shown in Figure 5. In large parts of the world, the population distance to water is, on average, similar to the dw_land, i.e. the dwr_pop/land is close to 1 (roughly one quarter of the data fall below a threshold of 0.8 while the median is 0.88; see Supporting Information S2). For many arid areas, however, the ratio dwr_pop/land is relatively low (Supporting Information S2); these areas include Australia, the Sahara, and Central Asia. On average, the populations in these areas live much closer to water than the average dw_land (Figure 5).

We also calculated the regional ratios of dwr_pop/land per climate zone (Table 4; Supporting Information S2). Again, we see that the ratio is lowest in arid zones, except for in Southeastern Asia, where populations in temperate zones live closest to water (Table 4). For half of the regions, the ratio is highest in the temperate zone, while for others it is highest in either the cold or tropical zones (Table 4). In many regions the differences are, however, rather small. At the regional scale, ratio of dwr_pop/land was smallest in Asia and largest in North America (Table 4; Supporting Information S2).

In order to examine statistical relationships at the FPU scale between population density and physical characteristics, we performed bivariate and multiple regressions using the SPSS software for the variables shown in Table 5. Data on precipitation and temperature were taken from the WorldClim v1.4 database [38], and refer to mean annual values for the period 1960–1990. We also used the aridity index of CGIAR [39]; this index represents the ratio of mean annual precipitation over mean annual potential evapotranspiration. The regression results are shown for the globe and per climate zone in Table 5. Bivariate regression results between all parameters (on a global scale) and multiple regression analysis for different arid regions are presented in Supporting Information S2.

On a global scale, we found significant bivariate correlations between population density and both aridity and precipitation (Table 5), indicating higher population densities with higher precipitation and lower aridity. However, when performing the bivariate regressions for each climate zone individually, the only significant correlations are in the cold region, for the parameters precipitation and temperature (Table 5). Similar results were found when performing multivariate regressions using two parameters. At the global scale, all combinations of parameters are significant, but within climatic regions significant regressions were mainly found in the cold region. The only exception is the combination of dw_land and precipitation, which resulted in a significant regression in arid zones.

Performing regression analyses using three parameters resulted in more interesting results. In the arid zone, adding dw_land to both precipitation & temperature and to aridity & temperature resulted in significant correlation, whereas there was no significant correlation between population density and the latter pairs of variables without dw_land. This indicates that population densities in arid zones are influenced by a combination of distance to freshwater bodies and precipitation/aridity. We also divided the arid zone into five geographical regions (see Supporting Information S2) and performed the same regression analyses as presented above, in order to find possible regional differences within the arid zone. We found that the correlations between dw_land and population density are strongest in Northern Africa and Middle and Southern Africa (see all the results in Supporting Information S2).

Overall, it seems that in the tropical and temperate zones the concentration of populations cannot be explained by either climatic factors or the distance to freshwater bodies. In the cold zone, climate variables play a very important role, whilst in arid regions population densities can be explained by a combination of climatic factors and distance to freshwater bodies.

We compared our results of population distance to water per FPU with estimates of water availability per person (in the year 2005) from Kummu et al. [17]. Figure 6 shows for each FPU the water availability versus the median population distance to water. The figure is divided into nine parts of a matrix. FPUs in the lower right corner are those which suffer from both chronic water shortage and for which the average distance to freshwater bodies is large. Almost all of the FPUs found in this part of the matrix are located in arid climate zones. However, not every arid FPU with a long distance to water suffers from water shortage, as can be seen from the points in the upper right corner. In contrast, there are also FPUs that suffer from chronic water shortage whilst having a relatively low population distance to water. These are in the lower left corner and are mainly areas with high population density in parts of Europe, East Asia, and South Asia (see Supporting Information S2).

A long distance to freshwater might be an extra stress factor on top of physical water shortage for populations living in such areas. Around 70% of the population under chronic water shortage (<1000 m³/capita/yr) lives in areas relatively close to water (<3.0 km), while 260 million people live in areas relatively far from water (>6 km), mostly in the arid zones of Middle East and Northern Africa (Figure 6). Approximately 20% of the global population lives in areas under some kind of water shortage (<1700 m³/capita/yr) and further than 3.0 km (global median) from the nearest freshwater body.