Quantifying Reservoir Outflow Performance in PCR-GLOBWB 2 with the Implementation of the Downstream Demand Allocation Function
Summary
During the 20th century, anthropogenic reservoirs have been constructed to ensure flood protection and an increasing global water demand. Consequently, natural streamflow timing and streamflow have been altered significantly. Increasing the anthropogenic impact on the hydrological cycle. To understand the feedback between artificial water management and global scale hydrological processes, Global Hydrological Models (GHM) are developed. In this study, a demand allocation function is integrated into the second version of PCRaster GLOBal Water Balance model (PCR-GLOBWB2), which fully integrates water demand at each time step. PCR-GLOBWB2 includes approximately 7000 human-made reservoirs, which are dynamically included according to their construction year, available in the most recently published Global Reservoir and Dam (GRanD v1.3) database. By integrating the allocation function it allows the user to study the reliability of reservoirs to provide sufficient release for downstream demand. Initially, downstream reservoir demand was limited to environmental flow, while the integrated function implements irrigational, domestic, industrial, and livestock demand as an addition.
The performance of PCR-GLOBWBs reservoir scheme and the integrated function were tested for 40 globally distributed reservoirs. Performance is validated and compared using output of four model simulations and observed outflow data for a time-range of 31 years (1980-2010). Simulation products include discharge for natural conditions, the implementation of reservoirs, circumstances including reservoirs and initial demand settings, and a combination of reservoir availability and the integrated demand allocation function.
Error metrics were given for each reservoir on a monthly basis. For twelve reservoirs the Kling Gupta Efficiency (KGE) was positive (>0.0), with maximum performance obtained for Ghost (0.64) and American Falls (0.64). Trends for the components of KGE obtained overestimations for the bias (>1.0, 27/40 reservoirs) and peak values (>1.0, 29/40), and a relatively well performing correlation (>0.5, 18/40). Highest performance trends were predominantly obtained for hydropower and within-year reservoirs, for which the average residence time is less than a year.
The ability of a reservoir to satisfy downstream demand was quantified in cumulative number of months with unmet demand. The allocation derived an average increase of approximately 65 months between the model simulations with environmental flow and allocated demand. Trends for the unmet demand obtained low sensitivity for hydropower and within-year reservoirs, while high sensitivities were obtained for non-hydropower and multi-year reservoirs, for which the average residence time is more than a year.
Both performance and unmet demand were related to the quality of meteorological forcing data. More accurate release modelled is obtained for reservoirs located in more accurately forced river basins. In conclusion, the reservoir scheme resulted in a relatively moderate performance, and the allocation function performed a more realistic representation of the downstream reservoir demand.