Estimation of missing precipitation records integrating surface interpolation techniques and spatio-temporal association rules

Abstract

Deterministic and stochastic weighting methods are the most frequently used methods for estimating missing rainfall values. These methods may not always provide accurate estimates due to their inability to completely characterize the spatial and temporal variability of rainfall. A new association rule mining (ARM) based spatial interpolation approach is proposed, developed and investigated in the current study to estimate missing precipitation values at a gauging station. As an integrated approach this methodology combines the power of data mining techniques and spatial interpolation approaches. Data mining concepts are used to extract and formulate rules based on spatial and temporal associations among observed precipitation data series. The rules are then used to improve the precipitation estimates obtained from spatial interpolation methods. A stochastic spatial interpolation technique and three deterministic weighting methods are used as interpolation methods in the current study. Historical daily precipitation data obtained from 15 rain gauging stations from a temperate climatic region (Kentucky, USA) are used to test this approach and derive conclusions about its efficacy for estimating missing precipitation data. Results suggest that the use of association rule mining in conjunction with a spatial interpolation technique can improve the precipitation estimates.