On the interpolation of volumetric water content in research catchments

作者：Dlamini Phesheya^*; Chaplot Vincent

来源：Physics and Chemistry of the Earth, 2012, 50-52: 165-174.

DOI：10.1016/j.pce.2012.09.008

• Interpolation techniques
• Geostatistics
• Secondary variable
• Volumetric water content
• South Africa

Digital Soil Mapping (DSM) is widely used in the environmental sciences because of its accuracy and efficiency in producing soil maps compared to the traditional soil mapping. Numerous studies have investigated how the sampling density and the interpolation process of data points affect the prediction quality. While, the interpolation process is straight forward for primary attributes such as soil gravimetric water content (theta(g)) and soil bulk density (rho(b)), the DSM of volumetric water content (theta(v)), the product of theta(g) by rho(b). may either involve direct interpolations of theta(v) (approach 1) or independent interpolation of rho(b) and theta(g) data points and subsequent multiplication of rho(b) and theta(g) maps (approach 2). The main objective of this study was to compare the accuracy of these two mapping approaches for theta(v). A 23 ha grassland catchment in KwaZulu-Natal, South Africa was selected for this study. A total of 317 data points were randomly selected and sampled during the dry season in the topsoil (0-0.05 m) for theta(g) by rho(b) estimation. Data points were interpolated following approaches 1 and 2, and using inverse distance weighting with 3 or 12 neighboring points (IDW3; IDW12), regular spline with tension (RST) and ordinary kriging (OK). Based on an independent validation set of 70 data points, OK was the best interpolator for rho(b) (mean absolute error, MAE of 0.081 g cm(-3)), while theta(g) was best estimated using IDW12 (MAE = 1.697%) and theta(v) by IDW3 (MAE = 1.814%). It was found that approach 1 underestimated theta(v). Approach 2 tended to overestimate theta(v), but reduced the prediction bias by an average of 37% and only improved the prediction accuracy by 1.3% compared to approach 1. Such a great benefit of approach 2 (i.e., the subsequent multiplication of interpolated maps of primary variables) was unexpected considering that a higher sampling density (similar to 14 data point ha(-1) in the present study) tends to minimize the differences between interpolations techniques and approaches. In the context of much lower sampling densities, as generally encountered in environmental studies, one can thus expect approach 2 to yield significantly greater accuracy than approach 1. This approach 2 seems promising and can be further tested for DSM of other secondary variables.

• 出版日期2012