Seasonal forecasting of reservoir inflows in data sparse regions
2018-06-22T13:31:11Z (GMT) by
Management of large, transboundary river systems can be politically and strategically problematic. Accurate flow forecasting based on public domain data offers the potential for improved resource allocation and infrastructure management. This study investigates the scope for reservoir inflow forecasting in data sparse regions using public domain information. Four strategically important headwater reservoirs in Central Asia are used to pilot forecasting methodologies (Toktogul, Andijan and Kayrakkum in Kyrgyzstan and Nurek in Tajikistan). Two approaches are developed. First, statistical forecasting of monthly inflow is undertaken using relationships with satellite precipitation estimates as well as reanalysis precipitation and temperature products. Second, mean summer inflows to reservoirs are conditioned on the tercile of preceding winter large scale climate modes (El Niño Southern Oscillation, North Atlantic Oscillation, or Indian Ocean Dipole). The transferability of both approaches is evaluated through implementation to a basin in Morocco. A methodology for operationalising seasonal forecasts of inflows to Nurek reservoir in Tajikistan is also presented. The statistical models outperformed the long-term average mean monthly inflows into Toktogul and Andijan reservoirs at lead times of 1-4 months using operationally available predictors. Stratifying models to forecast monthly inflows for only summer months (April-September) improved skill over long term average mean monthly inflows. Individual months Niño 3.4 during October-January were significantly (p < 0.01) correlated to following mean summer inflows Toktogul, Andijan and Nurek reservoirs during the period 1941-1980. Significant differences (p < 0.01) occurred in summer inflows into all reservoirs following opposing phases of winter Niño 3.4 during the period 1941-1980. Over the period 1941-2016 (1993-1999 missing), there exists only a 22% chance of positive summer inflow anomalies into Nurek reservoir following November-December La Niña conditions. Cross validated model skill assessed using the Heidke Hit Proportion outperforms chance, with a hit rate of 51-59% depending upon the period of record used. This climate mode forecasting approach could be extended to natural hazards (e.g. avalanches and mudflows) or to facilitate regional electricity hedging (between neighbouring countries experiencing reduced/increased demand). Further research is needed to evaluate the potential for forecasting winter energy demand, potentially reducing the impact of winter energy crises across the region.