Reconstructing missing daily precipitation data using regression trees and artificial neural networks for SWAT streamflow simulation

0022-1694/$ see front matter 2010 Elsevier B.V. A doi:10.1016/j.jhydrol.2010.09.005 ⇑ Corresponding author. Present address: Radioacti opment Division, Korea Atomic Energy Research Ins Yuseong-gu, Daejeon 305–353, Republic of Korea. Tel 42 868 8850. E-mail address: [email protected] (J.-W. Kim). Environmental models typically require a complete time series of meteorological inputs, thus reconstructing missing data is a key issue in the functionality of such physical models. The objective of this work was to develop a new technique to reconstruct missing daily precipitation data in the central part of Chesapeake Bay Watershed using a two-step reconstruction method (RT + ANN) that employed artificial neural networks (ANN) with inputs only from stations that were found to be influential in bootstrap applications of regression trees (RT). The predictive performance of RT + ANN was also compared with those of stand-alone RT and ANN methods. In addition to statistical comparisons of the reconstructed precipitation time series, these resulting data in the Soil and Water Assessment Tool (SWAT) watershed model were used to perform an error propagation analysis in streamflow simulations. The RT provided a transparent visual representation of the similarity between the stations in their daily precipitation time series. Seven years of data from 39 weather stations showed that both RT and ANN provided the reconstruction accuracy comparable to (or better than) published earlier results of precipitation reconstruction. The RT + ANN method significantly improved accuracy and was more robust when compared to RT or ANN methods. This method also provided more accurate and robust SWAT streamflow predictions with reconstructed precipitation. 2010 Elsevier B.V. All rights reserved.