Sørensen, P.B. & Johansson, L.S. 2020. Undersøgelse for mulige sammenhænge mellem miljøfarlige forurenende stoffer og oplandskarakteristika i vandløb. Aarhus Universitet, DCE – Nationalt Center for Miljø og Energi, 48 s. - Teknisk rapport nr. 177. http://dce2.au.dk/pub/TR177.pdf
The purpose of the project is to use existing data to uncover quantitative relationships between land use and measured concentration levels in running waters for the sediment and water phase, respectively, of environmentally hazardous substances (EHS) in streams. Data from both the NOVANA operational monitoring and surveillance monitoring have been included.
Sediment data from a total of 129 stations and water data from 154 stations are included. Each station was sampled between one and four times during the period 2011-2018. Water samples were typically gathered monthly each year. Where a station was sampled for several years, it is not necessarily the same substances that were analyzed every year.
The occurrence of EHS has been compared with the characteristics of the individual catchments of the monitoring stations as to the following attributes: Area fraction of Road, Urban Centre and Industry (AVBI), Area fraction of Temporary and Intensive Agriculture (AMIL) and catchment type where a distinction is made between the impact of the catchment in the form of Rain-Related Discharges (RBU) or Agriculture and Scattered Dwellings (LSB).
Primarily, a multivariate analysis was used following the same principle as that used in Sørensen and Johansson (2020). Methods include two steps for each combination of substance and sampling from, respectively, the sediment and the water phase. The first step investigates whether there are any correlations between the individual substances, expressed through principal components from a principal component analysis (PCA). The second phase uses the principal components scoring in a linear model to see whether a correlation occurs for either AVBI, AMIL or catchment type.
In general, there are only a few substances in the data material for which the measurements suffice for a PCA. Therefore, the analysis is carried out with a limited number of substances. It is important to point out that the significance levels obtained and expressed through the Pr value cannot be conventionally interpreted as the probability of obtaining the specific result assuming that the H0 hypothesis is correct. This is because the individual correlations are chosen among many possible and based on a low Pr value. The value should only be taken as an expression of which correlations are most likely to reflect a real relationship. Thus, this study alone uncovers those relationships that can be regarded as real relationships with the least uncertainty.
Perfluorinated compounds have been examined in some of the samples, but in many samples the concentration is measured to be below the detection limit and two different detection limits for the same substance have been used. Thus, for the substance, perfluorooctane sulfonate (PFOS), a left censored zero-inflated model was used, which directly attempts to model the conditions under which the sample was measured.
For the sediment samples, only substances in the group of PAHs met the criteria for statistical analysis. There was very strong covariation between the individual PAHs. Three PAHs were found in all sediment samples analysed for this substance group, and there was a significant positive relationship with both AVBI and catchment type. The relationship with catchment type also showed that RBU in particular could be source of PAH supply, i.e. overlap with AVBI.
The metals in the water samples exhibited substantial individual variation, indicating different sources and fates in the environment. A variation mostly related to the occurrence of nickel and absence of cadmium demonstrated a seasonal relationship, with the highest value in summer. One possible explanation may be that the summer water flow, due to a larger proportion of groundwater and waste water in the base flow, concentrates nickel, but dilutes cadmium, while it is the reverse for the winter water flow where the supply is more dominated by surface flow. The same variation also showed a positive correlation with AVBI. In a multiple metal analysis, both nickel and zinc showed the same tendency to correlate positively with AVBI and negatively with AMIL; in contrast, cadmium, chromium and, to a lesser extent, barium demonstrated a tendency to correlate negatively with AVBI. Copper and lead showed a somewhat unclear result without any clear correlation with AVBI. Point sources in the catchment showed a tendency towards a positive correlation with zinc, copper and lead and a negative correlation with cadmium and chromium.
Glyphosate and AMPA are the two most frequently found pesticides in the water samples and they exhibited a high degree of co-variation. However, there was also some variation where the two substances did not coincide due to different histories of degradation and transport. Coincidentally, glyphosate and AMPA exhibited seasonal variation with the highest concentrations in summer, which indicates that the base flow of the streams contains the highest concentrations of substances. Furthermore, there was a tendency for glyphosate, but not AMPA, to correlate with AMIL. A study of a slightly larger group of pesticides showed a weak positive correlation between MCPA and AMIL, while trichloroacetic acid and 2,6-dichlorobenzamide demonstrate a weak negative correlation.
Many perfluorinated compounds were below the detection limit and PCA was therefore only used for three substances. The analysis showed that the substances showed some co-variation among themselves, but also considerable individual variation. The zero-inflated left censored model described above was used to investigate the relationship between perfluorooctane sulfonate (PFOS) and AVBI. The result indicated a growing expected concentration with increasing AVBI, but no strong relationship, as only a small part of the variation was described. At the same time, the model estimated that approx. 20% of the samples collected from catchments largely without towns, industry and roads were samples that can be considered as having no substances (0 samples); the proportion of such 0 samples dropped to zero with an area share of about 25% for AVBI.
The relationships revealed in this statistical analysis can be used to guide the prioritisation of efforts in connection with establishing an actual modelling of concentration levels in these catchments. However, it should be pointed out that it was not possible to find close relationships between the catchment-specific independent variables (AVBI, AMIL and catchment type) and occurrence of substances, which indicates that it can be challenging to set up a model for predicting concentration levels based on land use and catchment type.
