Impacts of recent drought and warm years on water resources and electricity supply worldwide

We used a coupled hydrological-electricity modelling framework consisting of a hydrological model, stream temperature model, hydropower and thermoelectric power models, which is described in van Vliet et al (2016). This modelling framework focuses on the physical impacts of water constraints under climate variability on power plant usable capacities. Impacts of land use changes and economic feedbacks of the assessed water constraints (e.g. supply demands portfolio, electricity prices) are not modelled. The hydrological component consists of the variable infiltration capacity (VIC) model (Liang et al 1994), which is a grid-based macro-scale hydrological model that solves both the surface energy and water balances. The reservoir scheme of Haddeland et al (2006) which is combined with the routing model of Lohmann et al (1998) was used to include reservoir impacts on streamflow. This scheme assumes that reservoir operations follow simple operational rules based on the purpose(s) of dams (Haddeland et al 2006). The water temperature component of our modelling framework consist of the physically based stream temperature river basin model (RBM), which solves the 1D-heat advection equation (Yearsley 2009, 2012). RBM was previously modified for application on a worldwide level and to include effects of heat effluents from thermoelectric power plants and reservoir impacts on water temperature (van Vliet et al 2012a, 2012b).

Hydropower usable capacity for each hydropower plant was simulated based on regulated streamflow simulations from VIC, hydraulic head, total efficiency of the power generating unit, density of fresh water and the gravitational acceleration (see supplementary section 1). Thermoelectric power usable capacity was quantified using the model of Koch and Vögele (2009), which was modified as described in van Vliet et al (2012b). This model simulates the required water demand for cooling and usable capacity based on, amongst others, simulated streamflow and water temperature at the thermoelectric power plant site, both the total efficiency and electric efficiency, various parameters related to the cooling system type and environmental limitations for cooling water use (i.e. maximum river temperature (increase) and fraction of streamflow to be withdrawn). The model equations for calculating hydropower usable capacity and thermoelectric power usable capacity are presented in the supplementary section 1. For details of the coupled modelling framework and input datasets we refer to the supplementary information of van Vliet et al (2016).

We used the global Watch Forcing Data Era Interim meteorological dataset (Weedon et al 2014) as input into the VIC hydrological model and RBM stream temperature model to produce simulations of daily streamflow and water temperature on 0.5° × 0.5° spatial resolution worldwide for the period of 1979–2010 (including two years spin-up period). The modelling estimates of streamflow and water temperature were evaluated using observed station records of daily streamflow and water temperature worldwide, which showed a realistic representation of the observed conditions (see supplementary information of van Vliet et al (2016) for a general overview of the model performance and supplementary figure S1 for more detailed results for selected stations).

We applied the hydropower and thermoelectric power models on the plant individual level and extracted daily simulated streamflow and water temperature for 1981–2010 for the grid cell in which each power plant is situated. An evaluation of the impacts of biases (uncertainties) in simulated streamflow and water temperature on usable power plant capacity showed moderated impacts for most of the hydropower plants and thermoelectric power plants included in the uncertainty analysis (see supplementary section 2.3, supplementary table S1 and supplementary figures S3–S5).

We obtained information of power plant characteristics from the World Electric Power Plant Database (WEPPD) version 2013 (UDI 2013). The latitude and longitude of each power plant, hydraulic head for hydropower plants and water temperature limitations for thermoelectric power plants were derived as described in the supplementary information of van Vliet et al (2016) along with a quantification of the uncertainties in deriving these parameter estimates. For hydropower, we focused on conventional plants. For thermoelectric power we selected plants with availability of information on cooling system type, installed capacity, source of fuel, and use of river water for cooling. In total 24 515 hydropower plants and 1427 thermoelectric power plants are included (figure 1), which contribute to 78% of the installed hydropower capacity and 28% of the installed thermoelectric power capacity worldwide as reported by the Energy Information Administration (EIA accessed 2015). This smaller percentage for thermoelectric power is because more information for power plant characteristics (e.g. related to cooling system) is needed to parametrise the thermoelectric power model than for the hydropower model. The geographic distribution and age distribution of the selected plants was found to be representative for the full thermoelectric and hydropower power plant portfolio (see supplementary information of van Vliet et al 2016). Simulations of hydropower and thermoelectric power were previously evaluated using observed records on the country level for 1981–2010 (EIA accessed 2015). This showed that the interannual variability was slightly overestimated for some African countries, but for most countries worldwide the observed trends were realistically represented (see supplementary information of van Vliet et al (2016) for a general overview of the model performance and supplementary figure S2 for simulated and observed time series of selected countries).

Zoom In Zoom Out Reset image size

We used the gridded simulations of streamflow and water temperature to identify critical years with streamflow drought and high water temperature over 1981–2010. We quantified on a gridcell basis the mean number of days per year with streamflow smaller than the 10-percentile streamflow value, which is a widely used index for extreme low flow (Smakhtin 2001, Tharme 2003, Pyrce 2004). For water temperature, we quantified the number of days per year with values higher than the 90-percentile water temperature over the period 1981–2010. In addition, we calculated the number of days that both low flow and high water temperature coincide. As streamflow droughts and high water temperature events overall occur on a smaller geographic level than continental scale, we aggregated results over all gridcells per subregion to identify critical drought and warm years. We used the subregions defined by Giorgi and Francisco (2000) and adapted by Sheffield and Wood (2007). We focused our analyses of impacts of recent drought and warm years on 12 subregions with the highest number of hydropower and thermoelectric power plants and installed capacity according to the used plant dataset (figure 1). These 12 subregions together comprise 90% of the hydropower plants (n = 22 141) and 93% of the thermoelectric power plants (n = 1321) of the plant dataset.

For hydropower we identified recent critical years by selecting the three years with the highest number of days with streamflow drought in each subregion. In a similar way, we identified critical years for thermoelectric power by selecting the three years with highest number of days that both streamflow drought and high water temperature coincided. We focused on the three most critical years to show the impacts of drought and warm years with an average reoccurrence of about once every ten years in each subregion, and to compare impacts on this reoccurrence level between different subregions. In a next step, we quantified to what extent the usable capacity of all power plants per subregion deviated during these critical drought years compared to the average situation for 1981–2010. To eliminate the impacts of increasing trends in usable capacity due to increases in installed capacity over 1981–2010, we calculated annual average utilisation rates by dividing simulated usable capacity by installed capacity of each power plant and for each year. These utilisation rates of hydropower and thermoelectric power were subsequently aggregated over all power plants per subregion. Non-parametric Wilcoxon rank sum tests (Wilcoxon 1945) were then used to quantify to what extent hydropower and thermoelectric power utilisation rates deviated significantly during these drought and warm years compared to the average for 1981–2010. In addition, we quantified impacts of drought and warm years on plants at the individual level to explore the spatial variability in response of hydropower and thermoelectric power plants within the same subregion.

Annual time series with the mean number of days with streamflow drought and high water temperature are shown for the 12 subregions with the largest number of hydropower and thermoelectric power plants (figure 2). Aggregating results over all gridcells per subregion (solids lines) or only over gridcells where hydropower and thermoelectric power plants are situated (dotted, dashed lines) had overall very limited impacts on the identification of the critical years with streamflow drought (for hydropower) and critical years with both streamflow drought and high water temperature (thermoelectric power).

Zoom In Zoom Out Reset image size

Most severe streamflow droughts for hydropower and thermoelectric power were detected in northern Europe (2003) and in southern Africa (1992) with almost 100 days of streamflow drought (i.e. values lower than the 10-percentile streamflow). Most severe high water temperature years were found for the Amazon region (1998, 2010) with more than 100 days of high water temperature (i.e. values higher than the 90-percentile water temperature) (figure 2).

Comparison of these time series shows that detected years with streamflow drought and high water temperature highly differ between subregions. However, some streamflow drought and high water temperature events exceed the boundaries of subregions during some critical years. This was found for example for 1988 in North America (WNA, CNA, ENA), 2003 in Europe (NEU, MED) and 2007 in the eastern United States (CNA, ENA) with large regions with streamflow drought for more than 100 days and high water temperature for more than 50 days (figure 3; left and middle panel). Boundaries of subregions were also exceeded for example for 2009 in Asia (CAS, EAS, SAS) and 2010 in South America (AMZ, SSA) with large areas of more than 100 days of high water temperature (figure 3; middle panel). Areas where streamflow drought and high water temperature coincided for more than 50 days are rather small and are mainly located in central Europe (2003), the most northern part of India (2009) and the northern part of South America (2010) (figure 3; right panel).

Zoom In Zoom Out Reset image size

Table 1 shows an overview of the three most severe streamflow drought years (for hydropower) and the three most severe years with streamflow drought and high water temperature (thermoelectric power) for each subregion for 1981–2010. While the critical years for hydropower and for thermoelectric power do not necessary coincide, we found for all subregions (except East Asia) at least one (out of the three) similar critical years for both hydropower and thermoelectric power (see years indicated in bold in table 1).

On a mean worldwide level, simulated hydropower utilisation rates were on average reduced by 5.2%, and thermoelectric power by 3.8% during the investigated drought years compared to the average for 1981–2010. These global average values were obtained by calculating the average plant utilisation rate of the three critical years compared to the long-term average utilisation rate for 1981–2010, and then averaging the results over all plants worldwide. Boxplots with the distributions of utilisation rates of all hydropower plants for the 12 subregions are shown in figure 4(a). Values of 1 indicate that a power plant works at full capacity (no constraints) while for instance a value of 0.8 indicate that the plant works at 80% of the maximum capacity. The boxplots show that in Central North America, Eastern North America and Northern Europe, impacts of water constraints on hydropower utilisation are lowest, while impacts in Australia and East Asia are overall largest. This is related to the drier conditions and larger variability in streamflow in these regions. The interannual variability in utilisation rates of hydropower is highest in Australia and Southern Africa, but this might also be partly due to the low number of hydropower power plants in Australia (n = 130) and Southern Africa (n = 181) compared to the other subregions (n > 800) over which the results are aggregated. In most subregions, hydropower utilisation rates were significantly reduced (p < 0.05) for two out of the three critical drought years. For Western North America, Northern Europe, Mediterranean, Central Asia and East Asia, the hydropower utilisation rates were reduced significantly for all three investigated drought years (table 1).

Zoom In Zoom Out Reset image size

Boxplots with utilisation rates of thermoelectric power are presented for the six subregions with highest number of thermoelectric power plants (figure 4(b)). Our results show overall lower utilisation rates than for hydropower, which is expected as thermoelectric power usable capacity can be constrained by more factors (i.e. streamflow drought and high water temperature) and benefits less from storage of water in reservoirs during low flow conditions than conventional hydropower plants. Although considerable impacts of water constraints on thermoelectric power utilisation were found, these reductions were statistically insignificant (p > 0.05) for most critical drought and warm years compared to the average of 1981–2010.

We found however strong significant (p < 0.01) impacts on both thermoelectric power and hydropower utilisation rates during major critical years, such as 2003 in Europe (Northern Europe and Mediterranean) and 2007 in Eastern North America. During the drought year of 2003 in Europe, simulated hydropower utilisation was significantly (p < 0.01) reduced by 6.6% and thermoelectric power by 4.7% compared to the average of 1981–2010. Although the absolute values of hydropower utilisation rates are distinctly lower for the Mediterranean than for Northern Europe (figure 4), the relative changes (percentage reductions) in usable capacity for 2003 compared to the long-term average of 1981–2010 were larger for most hydropower power plants and thermoelectric power plants in Northern Europe than in the Mediterranean region (figure 5(a)). This discrepancy is partly due to the fact that power plants in the Mediterranean are hampered by water constraints on a more regular basis, resulting in lower absolute values of plant utilisation rates and smaller relative changes in usable capacity for 2003 compared to the long-term average for 1981–2010. In addition, the wet winter and spring of 2003 in the Mediterranean region resulted in relatively higher plant utilisation rates which partly compensated for losses during summer and resulted in smaller reductions on annual average basis, while water availability and plant utilisation rates in Northern Europe were lower throughout most of the year of 2003 compared to the long-term average of 1981–2010. For the drought year of 2007 in Eastern North America significant (p < 0.01) reductions of 6.1% in hydropower and 9.0% in thermoelectric power utilisation rates were found. Largest declines were simulated for hydropower and thermoelectric power plants in the state Georgia (figure 5(b)) where largest reductions in streamflow occurred during the drought year of 2007 compared to the average of 1981–2010.

Zoom In Zoom Out Reset image size