Comparison of spatial interpolation methods for soil moisture in Green Stormwater Infrastructure

Knowledge about soil-physical variables of Green Stormwater Infrastructure (GSI) like soil moisture (θ) is essential to understanding their hydrologic and treatment performance. θ depends on many local factors and is subject to high spatial and temporal variability (Takagi and Lin, 2012; Yao et al. 2013; Nasta et al. 2018). Information about the spatially continuous data of θ can help to understand the hydrologic response and provide an input for initial conditions to improve hydrologic modelling results. A number of deterministic and probabilistic interpolation methods and tools are available today to model the spatial distribution of environmental parameters such as θ (Li and Heap 2011; Yao et al. 2013). The quality of the spatial interpolation, however, depends on sample size, sample distribution and correlation to various other factors, for example terrain profile or vegetation coverage and makes the selection of the appropriate method difficult. Six methods commonly applied to soil characteristics have been selected to interpolate data that has been retrieved from 16 time-domain reflectometers measuring θ in the upper 30 cm of a GSI-site ́s surface. As θ changes at each sampling point also vary over time and therefore change the coefficients of some interpolation methods, estimates were compared for each hour of a 24-hour rainfall event. This is especially relevant as GSI soils are not only subjected to rainfall but also to distinct lateral inflows from impervious areas. Cross-validation and common error calculations were used to assess the statistical performance of the results and identify a method with least errors.