TY - CONF

T1 - Some challenges in seasonal forecasting for large lakes and reservoirs

AU - Sene, Kevin

AU - Tych, Wlodzimierz

N1 - Conference code: BHS

PY - 2018/1/24

Y1 - 2018/1/24

N2 - For large lakes and reservoirs, the inherent time delays arising from storage influences should in principle facilitate the seasonal forecasting of outflows. However, when developing forecasting techniques, a number of methodological issues need to be considered relating to both calibration and real-time operation of the models used.Here we consider some key points using examples drawn from case studies in Africa. In particular, one consideration is that the dynamic response can be complex with water levels responding over a range of timescales which, in addition to seasonal effects, may include rapid rises following heavy rainfall and longer-term declines lasting many years. External factors such as the El-Niño-Southern Oscillation may also have an effect over timescales of several years. This non-stationary response can cause some difficulties in evaluating model performance requiring careful choice of the metrics used. For reservoirs, an additional complication is that considerable detective work is often required to reconstruct the operating rules that were followed in the past.For real-time use, to help to account for deficiencies in model performance there is the option of updating outputs using data assimilation and, due to the persistence in levels and outflows, it may be possible to use techniques more typical of short-range forecasting such as autoregressive approaches. External climate drivers - represented by climate indices - might also be brought in at this stage. Where monitoring networks are sparse, it may also be useful to formulate the water balance in terms of net inflows or index series, rather than attempting to estimate individual components such as the lake rainfall and evaporation. For some reservoirs, though, complex logical rules may be required to represent the likely operator response and, even in the simplest case, the nonlinear dynamics usually require an ensemble approach to capture the uncertainty in model outputs.

AB - For large lakes and reservoirs, the inherent time delays arising from storage influences should in principle facilitate the seasonal forecasting of outflows. However, when developing forecasting techniques, a number of methodological issues need to be considered relating to both calibration and real-time operation of the models used.Here we consider some key points using examples drawn from case studies in Africa. In particular, one consideration is that the dynamic response can be complex with water levels responding over a range of timescales which, in addition to seasonal effects, may include rapid rises following heavy rainfall and longer-term declines lasting many years. External factors such as the El-Niño-Southern Oscillation may also have an effect over timescales of several years. This non-stationary response can cause some difficulties in evaluating model performance requiring careful choice of the metrics used. For reservoirs, an additional complication is that considerable detective work is often required to reconstruct the operating rules that were followed in the past.For real-time use, to help to account for deficiencies in model performance there is the option of updating outputs using data assimilation and, due to the persistence in levels and outflows, it may be possible to use techniques more typical of short-range forecasting such as autoregressive approaches. External climate drivers - represented by climate indices - might also be brought in at this stage. Where monitoring networks are sparse, it may also be useful to formulate the water balance in terms of net inflows or index series, rather than attempting to estimate individual components such as the lake rainfall and evaporation. For some reservoirs, though, complex logical rules may be required to represent the likely operator response and, even in the simplest case, the nonlinear dynamics usually require an ensemble approach to capture the uncertainty in model outputs.

M3 - Abstract

T2 - Seasonal Forecasting: Meeting User Needs

Y2 - 24 January 2018 through 25 January 2018

ER -