Comparison and application of multiple methods for temporal interpolation of daily soil moisture

Six interpolation methods for infilling daily in situ soil moisture are evaluated at 36 Oklahoma Mesonet stations from 2000 to 2007. The performance of artificial neural network (ANN), coefficient of correlation weighting, inverse distance weighting, daily average replacement (DAR), ordinary kriging, and spatial regression are compared using a leave-one-out cross-validation procedure. These six methods are evaluated based on their accuracy as well as their computational complexity and applicability to the North American Soil Moisture Database. We conclude that the ANN and DAR methods are the most accurate. Both methods were applied to infill soil moisture for cross-correlation analysis. The analysis was used to examine the relationship between near-surface and deeper soil moisture layers. Peak cross correlations between the 5 and 25cm layers varied between sites, ranging from 0.62 to 0.95 with an overall site average of 0.78. The lag at which the highest correlation between the 5 and 25cm layers occurred ranged from 1 to 4days. The relationship between the near-surface and deep soil layers is strongly modulated by spatial patterns of precipitation.