Ensemble Forecasting through Evolutionary Computing and Data Assimilation: Application to Environmental Sciences pp. 135-160
Authors: (M. Kashif Gill, Mark Wigmosta, Andre Coleman, Lance Vail, WindLogics, Inc, St. Paul, Minnesota, and others)
Abstract: A distributed modeling system for short-term to seasonal streamflow forecasts with the ability to utilize daily remotely-sensed snow cover products and real-time streamflow and meteorology measurements is presented herein. The modeling framework employs the state-of-the-art data assimilation and evolutionary computing strategies to accurately forecast environmental variables i.e., streamflow. Spatial variability in watershed characteristics and meteorology is represented using a raster-based computational grid. Snow accumulation and melt, simplified soil water movement, and evapotranspiration are simulated in each computational unit. The model is run at a daily time-step with surface runoff and subsurface flow aggregated at the watershed scale. The model is parameterized using a multi-objective evolutionary computing scheme using Swarm Intelligence. This approach allows the model to be updated with spatial snow water equivalent from National Operational Hydrologic Remote Sensing Center’s (NOHRSC) Snow Data Assimilation (SNODAS) and observed streamflow using an ensemble Kalman-based data assimilation strategy that accounts for uncertainty in weather forecasts, model parameters, and observations used for updating. The daily model inflow forecasts for the Dworshak Reservoir in north-central Idaho are compared to observations. The April-July volumetric forecasts issued by the U.S. Army Corps of Engineers (USACE) for Water Years 2000 – 2007 are also compared with model forecasts. October 1 and March 1 volumetric forecasts are comparable to those issued by the USACE’s regression based method. An improvement in March 1 forecasts is shown by pruning the initial ensemble set based on their similarity with the observed meteorology. The short-term (one-, three-, and seven-day) forecasts using Kalmanassimilation of streamflow show excellent agreement with observations. The scheme shows great potential for the use of data assimilation in modeling streamflow and other environmental variables.