Geostatistical tools in contaminated site investigation

Volatile chlorinated hydrocarbons (VCHCs) are potent solvents and were used extensively in the past, for example in laundries or in metal processing plants. Unfortunately, the necessary degree of care was not always taken when handling these chemicals, which are hazardous to health. In practice, it is therefore not uncommon in many places for the liquid CFCs to have leaked into the subsoil. There they can adhere to soil particles, run into the groundwater or evaporate and concentrate in the soil air. CFC spills are an example of harmful interference with the subsoil, where it is often necessary to carry out measures to remove the pollutants, for example by excavating contaminated soil material with a dredger and disposing of it. In order to be able to optimally use the means available for a remediation measure, the precise localisation of the contaminated area is of great importance. If you simply dredge on spec, you waste resources that could have been better used elsewhere. In order to determine the position of the damage in the subsoil, boreholes are usually drilled. Samples can be taken from the cores obtained, which can then also be assigned to a specific depth. Modern laboratory methods achieve detection limits of one part pollutant per million parts soil (and better!). However, the number of available cores and analyses is often limited. Obstacles such as buildings can also limit the areas of damage accessible to direct sampling. A complete analysis of every cubic centimetre of soil is neither practical nor useful. Nevertheless, it is helpful for the processing of the damage case to draw conclusions from the point information from the soil samples to the entire volume of the pollutant distribution. Geostatistical methods can help with this task. They establish a mathematical relationship between the measured concentrations and their respective position in space. With the help of such geostatistical models, interpolations can be made on the basis of the available measurement data, so that a probable concentration is calculated for each point in space. Geoscientists have an ever-growing canon of powerful algorithms at their disposal for this task. In addition to powerful classics such as Kriging and Generalised Additive Models, tools from the group of so-called machine learning have increasingly been added in recent years. Random Forest and Support Vector Machine models as well as Artificial Neural Networks can be adapted and used to solve situation-related problems. The result is a continuous, n-dimensional image of the pollutant distribution. From this, focus areas can be derived, for example where the concentration is particularly high or where there is a risk of endangering a sensitive protected good. Geostatistical methods help us to better quantify the effects and chances of success of a remediation measure in advance. In this way, measures can be precisely adjusted so that they have the maximum effect in the terrain.

#engineeringforabettertomorrow, #mullandpartner, #whatwedo, #groundwaterprotection, #rstats, #hydrology #mupgroup