October 8, 2018

Reservoir Inflow Forecasting

Posted by Jason Lillywhite
Last month, at the annual symposium hosted by the Arizona Hydrologic Society, I presented on probabilistic reservoir forecasting using GoldSim. This model combines many existing components available in our library to forecast snowmelt driven runoff inflows to a reservoir and estimates risk of spills and/or shortages.

This model demonstrates the use of multiple GoldSim model components pulled from our library to forecast inflows to a reservoir while incorporating risk and uncertainty. The model is applied to a reservoir located in the mountains east of Salt Lake City, Utah and is calibrated to over 20 years of historic time series of snowpack and streamflow data. This model predicts the risk of flood due to uncontrolled flows over a spillway and/or water supply shortages in drought years. In the first 14 days of the simulation, it relies on a 14-day weather forecast then transitions to a probabilistic weather generator after the 14 days. The model includes snow accumulation and melt, watershed runoff and routing, and reservoir operations.

Model Calibration

First the model was calibrated to ensure that the logic in the model is robust enough to represent the physical system under a variety of conditions.

Forecast Model

Once the model was sufficiently calibrated, the following components were added to the model:

  • Weather forecast (14-day forecast for Salt Lake City, then correlated to the site)
  • Probabilistic, Markov-chain weather generator (for simulation time after the first 14 days)
  • Uncertainty for static and dynamic inputs (watershed characteristics, climate factors, etc.)
  • Monte Carlo simulation
  • Estimation of risk of not meeting water supplies and also for uncontrolled spills due to the water level increasing to a level above the spillway crest.
Below is a screen capture of the model interface: 

If you would like to download a copy of this model and try it out yourself, please visit our Model Library, here.

No comments:

Post a Comment