Sampling and analytical plus subsampling variance components for five soil indicators observed at regional scale

When comparing soil baseline measurements with resampled values there are four main sources of error. These are: i) location (errors in relocating the sample site), ii) sampling errors (representing the site with a sample of material) iii) subsampling error (selecting material for analysis) and iv) analytical error (error in laboratory measurements). In general we cannot separate the subsampling and analytical sources of error (since we always analyse a different subsample of a specimen), so in this paper we combine these two sources into subsampling plus analytical error. More information is required on the relative magnitudes of location and sampling errors for the design of effective resampling strategies to monitor changes in soil indicators. Recently completed soil surveys of the UK with widely differing soils included a duplicate site and subsampling protocol to quantify ii), and the sum of iii) and iv) above. Sampling variances are estimated from measurements on duplicate samples -- two samples collected on a support of side length 20 m separated by a short distance (21 m). Analytical and subsampling variances are estimated from analyses of two subsamples from each duplicate site. After accounting for variation caused by region, parent material class and land use, we undertook a nested analysis of data from 196 duplicate sites across three regions to estimate the relative magnitude of medium-scale (between sites), sampling and subsampling plus analytical variance components, for five topsoil indicators: total metal concentrations of copper (Cu), nickel (Ni) and zinc (Zn), soil pH and soil organic carbon (SOC) content. The variance components for each indicator diminish by about an order of magnitude from medium-scale, to sampling, to analytical plus subsampling. Each of the three fixed effects (parent material, land use and region) were statistically significant for each of the five indicators. The most effective way to minimise the overall uncertainty of our observations at sample sites is to reduce the sampling variance.