Evaluation of sampling techniques to characterize topographically-dependent variability for soil moisture downscaling

- Soil moisture downscaling can be based on smaller datasets than previously used.
- Use of strategic sampling (SRS or cLHS) enhances the value of small datasets.
- SRS and cLHS each can perform best in particular situations.
- The EMT model outperforms the EOF method for downscaling with small datasets.

SummaryDownscaling methods have been proposed to estimate catchment-scale soil moisture patterns from coarser resolution patterns. These methods usually infer the fine-scale variability in soil moisture using variations in ancillary variables like topographic attributes that have relationships to soil moisture. Previously, such relationships have been observed in catchments using soil moisture observations taken on uniform grids at hundreds of locations on multiple dates, but collecting data in this manner limits the applicability of this approach. The objective of this paper is to evaluate the effectiveness of two strategic sampling techniques for characterizing the relationships between topographic attributes and soil moisture for the purpose of constraining downscaling methods. The strategic sampling methods are conditioned Latin hypercube sampling (cLHS) and stratified random sampling (SRS). Each sampling method is used to select a limited number of locations or dates for soil moisture monitoring at three catchments with detailed soil moisture datasets. These samples are then used to calibrate two available downscaling methods, and the effectiveness of the sampling methods is evaluated by the ability of the downscaling methods to reproduce the known soil moisture patterns. cLHS outperforms random sampling in almost every case considered. SRS usually performs better than cLHS when very few locations are sampled, but it can perform worse than random sampling for intermediate and large numbers of locations.