Estimating missing data of wind speeds using neural network

In a lake system, wind data is important for hydrodynamics and sediment transport modeling. However, there exists missing data caused by instrumental failure due to birds, thunderstorms, or other unexpected events. Missing data will degrade the performance of modeling approach and accuracy of model results. In order to overcome this problem, we have developed a neural network model that attempts to "learn" and "discover" wind speed behavior from available data and to estimate the missing data. By applying statistics and z-scored distribution coupled with multi-variable time lag analysis, the synthetic wind speeds for missing data are obtained. The results of this approach are better than those of the traditional nearest neighbor approach. Wind data collected from Lake Okeechobee, the second largest freshwater take within the United States, will be used as a test database. The developed model demonstrates its abilities to reproduce accurate wind speed for the years 1996 and 1999.