Intelligent Soil Sampling

The adaptive sampling scheme being put through its paces in the field. A sensor records soil moisture and a GPS antenna on top of the probe records its position. Both sets of information are fed to the hand-held computer which analyses the data and determines the next sampling site.

The complex nature of soil raises a number of questions that can influence decisions about the management of land, on both a national and local level. For example: Are the nation's soils losing carbon? Where do concentrations of lead exceed regulatory thresholds? Is the soil acidity in a particular field a limiting factor on crop yield? The problem is that the soil is so variable, even within a single field, that it is difficult to obtain this information efficiently.

The Environmetrics group at Rothamsted Research, led by Dr Murray Lark, have developed an intelligent computer program, with funding from a BBSRC Industrial Partnership Award with the Home-Grown Cereals Authority, which can tailor soil sampling to local conditions, allowing land managers to obtain high quality information, while avoiding costly over-sampling. This method has potential to improve the management of agricultural fields and for mapping soil pollution and monitoring the success of clean-up operations.

"Our program 'learns' about the variation of the soil as sampling proceeds, and is therefore able to generate a sampling scheme that is tailored to local conditions and ensures that the sampling effort is used to its greatest effect," explains Dr Lark.

The program's underlying concept is the variogram - a mathematical model of how soil varies across an area. As sampling begins, the computer program is 'ignorant' of the variogram and uses data from sampling to reduce the level of uncertainty and to direct where subsequent samples should be taken. As data accumulates this uncertainty is reduced. Once the program has a sufficiently robust model of the spatial variation within the area, a final phase of sampling points is identified to ensure that the definitive map of the soil will be sufficiently precise. Both computer simulations and practical trials have shown that this adaptive sampling scheme can converge from no initial knowledge to a reliable map of how the soil varies.

Interestingly, the program can tell the difference between uniform and complex soils (see illustrations). "We have tested the program with two simulated patterns of soil variation," says Dr Lark. "Our program rapidly identifies where variation is complex and many sampling sites are needed. But where variation is gradual, with large contrasting patches, the program recognizes that much less sampling is needed, and samples can be further apart."

When tested on real landscapes, the scheme has reduced the number of sampling sites needed, thus saving time and money, without any loss of accuracy. Several other potential applications of the technology are now being considered.

Two simulated patterns of variation of a soil property (top) and the optimized sampling schemes (bottom). On the left the variation is more intricate, dominated by components that are only spatially dependent over short distances. On the right spatial dependence is seen over longer distances.

Dr Lark was head of the Environmetrics group at the former Silsoe Research Institute. In 2004, the group transferred to the newly formed Biomathematics and Bioinformatics Department at Rothamsted Research.

Dr Murray Lark (murray.lark@rothamsted.ac.uk)

Back to Science Snapshots