Sampling scheme for mapping variability of soil hydraulic properties in an agricultural area in Tanzania

Management of water in cropping systems necessitates a characterisation of spatial variability of soil hydraulic properties (SHP) (water retention and hydraulic conductivity). Digital soil mapping (DSM) provides a toolset of statistical methods for mapping SHP over landscapes. Designing a spatial sample is an important first step in DSM. In this research, SHP will be mapped using both direct measurements of SHP and soil texture which is related to SHP. This research addresses the problem of designing an efficient sampling scheme for mapping SHP combining these two types of measurements. A stratified random sampling design was adopted. The strata are 100 compact geostrata of equal area constructed by k-means clustering of pixels constituting the study area, using the spatial coordinates of the pixels as classification variables. This method was implemented in the (add-on package spcosa of the statistical software R). One location was randomly selected from each geostratum generating 100 master locations. The 100 geostrata were then collapsed into 50 pairs of geostrata based on their geographical distance. Two neighbouring geostrata formed a pair, with two selected sampling locations. At one of these pairs, both SHP and soil texture were measured generating a sub-sample of 50 calibration locations. At the remaining 50, only soil texture was measured. Calibration locations data will be used to calibrate a regression model relating SHP to soil texture. This sampling design leads to an even spread of both the master locations and calibration locations across the study area, desirable for mapping with a spatial interpolation method.