Are data collected to support farm management suitable for monitoring soil indicators at the national scale?

摘要

Monitoring of topsoil properties (referred to as indicators) at the national scale has been limited in general to government-funded representative surveys. We consider a cost-effective complementary source of soil information for monitoring agricultural soil across England and Wales (E&W): soil measurements paid for by farmers that we refer to as farmers' data (FD). A potential problem in using FD for soil monitoring is any unattributable sources of bias, such as the sample design. Farmers may choose to focus their measurements (purposively) where they perceive a particular problem. Such a source of bias is avoided in the random sampling adopted by statistically designed surveys, such as the Countryside Survey (CS2007) and LUCAS (Land Use/Cover Area frame statistical Survey). We used measurements from 143 000 FD soil samples from a single laboratory to estimate national mean values and confidence intervals of five topsoil indicators (pH, available P (Olsen), K, Mg and organic matter (OM)) across three combinations of nation (England or Wales) and land use (arable and horticulture (A&H) or improved grassland (IG)). We computed mean estimates for FD over two time periods (2004-9 and 2010-2105) and assessed the significance of any change. We compared these estimates with those from representative national surveys to establish whether there was evidence for bias and whether it could be explained. Mean estimates of topsoil pH for the FD and the LUCAS survey (same analytical method) were consistent for both A&H and IG. Although FD estimates of mean Olsen P (OP) concentrations were similar to previous surveys, we show it is likely that the larger mean OP concentrations observed in the LUCAS survey compared with FD for arable topsoil in England are partly due to an attributable source of analytical bias. For such quantifiable sources of bias, it might be possible to adjust estimated mean values from FD. However, FD might also include sources of unattributable bias, such as the effect of purposive sampling. It is important that contemporaneous data from surveys with statistically unbiased designs are available so that we can assess whether unattributable sources exert a significant effect over estimates of mean values computed from FD.