New optimal proximity-based imputation, k-nn (k-nearest neighbor) classification and k-means clustering methods are proposed and developed for estimation of missing precipitation records in this study. Variants of these methods are embedded in optimization formulations to optimize the weighing schemes involving proximity measures. Ten different binary and real valued distance metrics are used as proximity measures. A temperate climate region, Kentucky in the United States, is used as case study to evaluate the efficacy of these methods for estimation of missing precipitation data. A comprehensive exercise is undertaken in this study to compare the performances of the developed new methods and their variants to those of already available methods in literature. Several deterministic and stochastic spatial interpolation methods and their improvised variants using optimization formulations are used for comparisons. Results from these comparisons indicate that the optimal proximity-based imputation, k-mean cluster-based and k-nn classification methods are competitive when combined with mathematical programming formulations and provided better estimates of missing precipitation data than available deterministic and stochastic interpolation methods.
展开▼
