Author(s): Ramesh S. V. Teegavarapu; Thu Nguyen
Linked Author(s):
Keywords: No Keywords
Abstract: Missing precipitation data is a major issue for many hydrological modeling studies requiring chronologically continuous and error-free data, including climate change and variability. Emerging machine-learning approaches are found to be beneficial for the estimation of missing data. In the current study, missing daily precipitation records are imputed using regression and model tree-based approaches. These approaches are used in a spatial interpolation mode as autocorrelation of the precipitation series at several lags is deemed weak for any temporal interpolation schemes. Missing data at a site is estimated using observations at other sites in a region using the tree-based approaches, which partition the data along with the development of multiple local models for imputation. Daily precipitation data at twenty-two sites in the state of Kentucky are used for the application of the tree-based approaches. Results from this study indicate that tree-based approaches can be used to impute missing precipitation data, with the model tree-based approach providing better estimates than the regression tree method.
Year: 2024