Urban growth, climate change, and freshwater availability

Demography.

Base demographic data on cities were taken from GRUMP (4, 31). GRUMP urban extents are spatially defined primarily on the basis of satellite imagery of night-time lights. The extents are then linked with information from censuses and gazetteers containing population information on hundreds of thousands of urban settlements, including those with quite small populations. The algorithm that defines urban extents, especially for large urban agglomerations, typically gathers contiguous urban and suburban areas into one urban extent. For our study, we evaluated the water availability and population for the entire urban extent as measured by GRUMP; this assumes water sharing among constituent municipalities that may or may not occur in practice.

Demographic projections for the urban areas to the year 2050 were taken from Balk et al. (5). For these cities, a time series of population at the city level was obtained from the United Nations (UN), using all information available; for most countries this means a series from 1970 onward. Panel-data regression models (with an allowance for city-specific features captured statistically through random or fixed effects) were used to estimate the historical drivers of city growth and to forecast city populations to 2050. We examined two sets of control variables in these regressions. The first set, producing what we term the Basic Demographic scenario, is based solely on national-level urban rates of fertility and child mortality, city size, and some correction factors to account for how the implicit spatial boundaries of a city have changed over time (using the UN's definitions of city proper, urban agglomeration, and metropolitan region). We augment these controls to produce the Ecological Factors scenario, adding multiple categorical variables for ecosystem (using definitions from the Millennium Ecosystem Assessment (32) and a low-elevation coastal zone (33). This scenario allows the population of cities in specific biomes (e.g., arid regions) to grow slower or faster to the extent that observed population dynamics have consistently correlated with biome in the past. Note that these demographic projections—which are attached to a spatial reference for 2000—do not include estimates of how urban spatial extents will change by 2050.

Hydrology.

The water balance calculations were carried out at 6' (longitude × latitude) spatial resolution using version 6.01 of the Simulated Topological Network (STN30p), a modestly updated version of the dataset presented in Vörösmarty et al. (6). The water balance/transport model applied in the present study (WBMplus) is an updated version of the global water balance model that was developed by Vörösmarty et al. (7, 8) and subsequently modified by Wisser et al. (9, 10). Input data on precipitation and temperature use are fed into the model, as well as a hydrologic network based on a digital elevation model. Evapotranspiration is calculated as a function of local land use, including consumptive water use by agriculture.

We use four scenarios of future hydrology, based on the four scenarios in the Millennium Ecosystem Assessment: Adaptive Management; Global Orchestration; Order from Strength; and Technogarden. More detail on the assumptions behind these four scenarios can be found in the volume of the Millennium Ecosystem Assessment on their scenarios (11). Importantly, each scenario represents a spatially explicit model of economic development, greenhouse gas (GHG) emissions, and land-use change, including agricultural expansion. For each of the four scenario's GHG emissions timeline, a climate model was used to simulate the climate in 2050. Changes in precipitation or temperature then impact the modeled hydrologic cycle. The variation in hydrology among our four scenarios is thus due to variation in the extent of land-use change and GHG emissions.

We have chosen to use the Millennium Ecosystem Assessment scenarios because of the spatially explicit and logically consistent forecasts of land-use change and GHG emissions embedded with them. However, the most recent GHG forecasts of the Intergovernmental Panel on Climate Change (IPCC) have changed somewhat. To give readers a sense of the assumptions underlying our scenarios, our four scenarios emissions pathways are shown next to the current IPCC scenarios (Fig. S2).

Analysis.

Using river network topology, we calculated the available water for each urban extent. Cities on small islands (e.g., Comoros Islands) that were not modeled in our hydrologic network are excluded from the analysis. Available water included local runoff generated within the city extent and water flowing into the city. In one sense, this is a very optimistic number: all water is assumed available for use, even water than ends up falling on rooftops or city streets. Moreover, we are assuming that water use among cities in the same watershed is nonconsumptive and that water can be reused several times as it flows down the river. Urban water use is generally nonconsumptive in this sense, except that water quality issues may make some water unusable for downstream users without expensive treatment plants. Our treatment of urban water use as nonconsumptive is a different perspective from many water availability indices calculated at a watershed level, whereby dividing watershed available water by watershed population water use is implicitly defined as consumptive.

Various standards have been used for the minimum amount of water for daily needs, depending on what needs are considered in the estimate (14). We define water shortage as <100 L per person per day, following the World Health Organization definition of “optimal” water for all domestic needs, including drinking, cooking, bathing, cleaning, and sanitation (15). If a city's average water availability is less than this standard it is defined as having perennially water shortage. For cities that did not satisfy their water requirements within their urban extents, a series of buffers was used in an iterative process (Fig. S1). First a buffer of 10 km was used, and if the water shortage standard was not achieved then buffers of 30 km, 60 km, and 100 km were used. Because the population of the urban extent did not change during this buffering process, available water per capita by necessity stays the same or increases as buffer distance increases.

Because our hydrologic model outputs data on a monthly time step, it is relatively easy to incorporate seasonal variability into our analysis. An urban extent is defined as having seasonal shortage if there is at least 1 mo per year in which it does not have 100 L per person per day. Note that many cities meet this criterion for seasonal shortage and may in practice avoid problems simply by having sufficient water storage to make it through the dry month or months.

To examine how urban water use might affect freshwater biodiversity, we intersected our map of cities with insufficient water and a map of the freshwater ecoregions of the world. Freshwater ecoregions are spatial areas of similar ecology, often in the same major hydrologic drainage or of similar geology. If a city with insufficient water is located in a freshwater ecoregion, this means there is at least one stream in the ecoregion that is being fully used by urban residents. When an urban extent or its buffered area of water acquisition crossed the boundary of two freshwater ecoregions, in our calculation the population of the city with insufficient water was partitioned between the two ecoregions according to the volume of water available within that portion of each ecoregion. Population with insufficient water by freshwater ecoregion was graphically related to freshwater fish richness and endemism, because freshwater fish are one of the taxa most likely to be affected by water withdrawal.