As part of planning the spreading of sludge from urban waste-water treatment plants since 1984, a large number of analyses of eight trace metals (TM) have been carried out on soil samples at national scale before spreading any sludge. ADEME requested INRA (in Orléans) to collect, computerize and process the available data. Owners of such analytical results were asked to provide, if possible, georeferenced data to INRA. The results were to be strictly confidential as the analysed plots will never be identifiable on the final reports and maps. Collecting and computerizing of the data was done in 1997 and 1998 at INRA Orléans and between 2003 and 2004 these data were reprocessed to obtain the present report. The created database is now located at ADEME and GIS Sol and provides new information on TM contents in French agricultural soils and their spatial distribution.

Chapter 1. Collection of data and related information

According AFNOR standard NF U 44.041 (used before 1998), the content of eight trace metals (Cd, Cr, Cu, Hg, Ni, Pb, Se, Zn) was determined in soil samples coming from surface horizons of agricultural land where sludges were supposed to be spread. All trace-element contents were obtained from the assays based on air-dried "fine earth" (i.e. the <2 mm fraction) samples and are expressed in mg.kg-1. Collected data also included results from “standard” agro-pedological characterizations as particle-size distribution (five fractions), pH in water, CEC, organic carbon and total lime.

If available data related to the analyses was also recorded, such as the digestion method, the identity of the laboratory, the date of the analyses. In addition, an effort was also made to collect as far as possible data on the sampling location (e.g. name of the municipality, geographical coordinates of the site based on the Lambert coordinate system), the name of the company that had carried out the sampling, land use and land management of the plot, soil type and/or nature of parent material.

The database covers:

• Only ploughed surface horizons (e.g. topsoils);

• Almost exclusively cultivated land that would receive waste-water treatment sludge;

• Samples collected before May 1998 and before any spreading of sludge. However, it remains possible that some samples were taken in plots where sludge had been spread many years ago before the present rules and regulations came into force;

• Analyses carried out in at least 28 different laboratories, using at least two different digestion methods before assaying;

As a result of such collection 11,396 samples are registered in the database.

Obviously, there is no true sampling strategy, as results came from the analyses made by dozens of different operators. As a consequence, the sampling does not faithfully represent all soils in France, but it can be assumed that the sampling is fairly representative of the tilled agricultural soils in the regions where the samples were taken.

No selection was made on the basis of a “normal” or “anomalous”, or even clearly contaminated, character of the measured TM contents. The database thus certainly contains:

• Samples whose TM concentration remains very close to the “natural pedogeochemical concentrations” (NPGC), which is doubtlessly the case for grassland that has been little or not fertilized;

• A large number of samples corresponding to "usual agricultural contents" = NPGC + notable input of agricultural origin and diffuse atmospheric deposition;

• Some samples that are notably or very strongly polluted.

During the study many practical difficulties were encountered, some related to the data-collecting itself, others to the heterogeneity of the collected data and information.

• Data were scattered over various institutes and located in diverse documents. Some were not easy to find and it was difficult to avoid doubles.

• In over 80% of the cases, the geographical location in Lambert coordinates was missing. Generally, the only documents allowing the georeferencing of the samples were copies of the topographical background maps at scales 1:25,000 or 1:50,000. It was thus necessary to check the coordinates of the sampling points or plots on topographical maps. Such manual input of geographical coordinates, whether carried out by the data suppliers before sending us the files, or later by us, was a very unwieldy operation and may be a source of errors.

• Several types of technical errors were identified: input errors, column or unit errors, quantities of metal extracted by EDTA confused with total contents, etc. Some could be corrected as the values were “outside the norm” and thus attracted our attention, but many others, similar but less obvious, probably remained unnoticed.

• In many cases, the data turned out to be incomplete or unusable. For instance, the particle-size distribution of the “fine earth” is commonly missing, even though this is quite useful information. In addition, many of the data concerning the pedological and geological context were inconclusive, heterogeneous and unreliable.

• Some of the older TM analyses (1984-1993) show very high contents that seem suspect. Their reliability is thus difficult to judge, but apparently some of the laboratories that progressively started assaying for TM encountered difficulties during the first months.

• Numerous results are mentioned as being “below the determination limit” (e.g. Cd < 0.50 mg.kg-1, or Hg < 0.30 mg.kg-1, or Se < 1.0 mg.kg-1). In many case, such information is unusable as an example, the median for cadmium content is 0.26 mg/kg, calculated over 10,634 nation-wide available values.

• Another source heterogeneity concerns the analytical methods used by different laboratories for determining TM contents, and mainly the digestion method before assaying. Most laboratories used aqua regia, but some used an association of two strong acids: HF + HClO4.

Chapter 2 Analytical methods – Consequences

Two methods are used for sample dissolution before assaying by the analytical laboratories. It is pointed out that “HF + HClO4 (hereafter “HF”) method” provides truly “total” values, whereas the “aqua regia method” (hereafter “ER”) gives “pseudo-total” values. In addition, the difference between these methods depends on the soil composition and on the considered element.

Chapter 3 Statistical processing of TM data

The statistical methods used to describe the TM concentration data are presented in this chapter together with essential notions (as the definition of “outlier values”). The following results are then proposed:
*   Statistics on all TM and agro-pedological data covering all France (see tables 3.2 and 3.3);
*   Information on a national scale concerning the number of “outlier values” and of values over the threshold values coming from the regulation of sludge disposal (see table 3.3);
*   Localization in each “département” of “outliers” and other high values (see maps 3.1 to 3.4);
*   Statistics made by distinguishing between “HF” and “ER” populations (see table 3.5 and 3.6);
*   Correlation matrix integrating all TM values on a national scale (correlations generally poor, as parent materials and soils are very diverse), including some examples made at the "département" scale with a generally better correlation (see tables 3.8 and figures 3.4 to 3.7).

Chapter 4 Statistical processing by textural classes

The all data was stratified into five textural classes presenting increasing clay contents (particles <2 µm). A strong statistical relationship clearly appears between TM contents, except Hg, and clay content. A method to search for TM contamination based on a grain size approach was applied to samples with a sandy texture.

Chapter 5 Statistical processing by parent material

Despite the absence of reliable data on parent materials and soil types, it was attempted to regroup all analyses coming from the same parent materials or “pedo-geological families”, by applying a statistical processing to define their main pedo-geochemical traits. As only surface horizons of cultivated soil are concerned, all these samples bear the mark of “diffuse contamination” (agricultural input, atmospheric fallout) of variable importance.

Only five pedo-geological families could be isolated and studied (see chapter 5.2):

• Soils of the Paris Basin developed in loamy materials (n = 1089);

• Soils developed from chalk in Champagne (n = 253);

• Sandy podzolic soils of the Landes de Gascogne (SW France) (n = 24);

• Soils developed in Alpine moraines (n = 189);

• Some sandy “boulbène” soils of the Ariège (n = 20).

Measured concentrations are highly contrasting, showing that the pedo-geochemical inheritance remains very obvious in the surface horizons, notwithstanding the diffuse contamination. The main types of parent material thus are an excellent basis for data stratification, but certain information, needed to classify the samples according to this criterion, is commonly missing.

Chapter 6 Statistics by "“département”

This chapter mainly presents statistical tables drawn up for the eight trace metals and for the 31 "“départements” for which at least 100 measurements were available. It was not possible to separate the data into the “HF” and “ER” sub-populations.

Chapter 7 Statistics by agricultural region (AR)

The available data were processed on a territorial basis of “agricultural regions” (AR). This territorial subdivision of France was made by geographers shortly after the last world war, aimed at creating homogeneous areas in terms of “soil nature” as well as of climatic conditions and the main type of agricultural vocation. Eight maps present the number of national outlier values by AR; another eight show the median values calculated for the AR with at least 21 usable measurements (see maps 7.1 and 7.2). Some ARs show a large number of outlier values, or median values much higher than the rest of France. Generally, this results from natural pedo-geological anomalies, but in some cases this reflects the impact of human activity. The sampling stratification by agricultural region seems much more useful than that by “département”.

Chapter 8 Spatializing of TM values

Several hundreds of available data for a “département” are still relatively few in order to perform a spatial mapping, and moreover such data are commonly unevenly distributed. In addition, no logical basis exists for extrapolating such values in space. The distribution of data in space is showed under the form of cartograms at regional scale (i.e. 8 areas x 8 metals = 64 cartograms).

However, geostatistical mappings for lead and nickel were tried for 9 “départements”, around Paris and in Normandy, for which relatively abundant data were available and rather well distributed. Standard kriging was used as interpolation method and three types of presentation were compared (number and subdivision of classes). Applied to such a database, kriging interpolation provided estimation “maps” that seem to be not very pertinent.

Chapter 9 In-depth work on the Yonne département

The objective was to complete as much as possible the data collected in 1998 to see what can be done with a larger number of data, especially with data on which the geological and pedological contexts are well known. The Yonne département was selected as it is a well known area both in terms of geology and pedology. It was possible to stratify the sampling by small natural regions (SNR) and by parent material, which showed great differences in metal concentrations from one SNR to the next, as well as between different parent materials (see figure 9.10 and map 9.8).

Chapter 10 Perspectives

Since 1998 the situation has changed in France, as today other analytical databases are available for TM in soils (e.g. RMQS database, www.gissol.fr). The question now is whether a new collection of TM analyses should be launched to integrate data obtained since 1998 and if possible also those omitted during the first collection.

If the answer is YES, then the imperfections of the first collection should be taken into account and corrected (if possible). Another way to operate may be to collect data directly from the analytical laboratories instead of asking the "final clients", such as engineering companies or governmental agencies.

As far as future statistical data-processing is concerned, two data-stratification methods can be envisaged:

- By parent material, provided ad-hoc information is available to assign each analysis in a reliable manner to a specific parent material, which is rarely the case;

- By agricultural region (AR) or small agricultural region (SAR), as georeferencing by commune suffices for knowing automatically in which AR or SAR the sample is located.

An ideal situation would be to be able to stratify the sampling by small natural regions (SNR) as this is the best territorial subdivision for taking into account rock types and parent materials, relief, and soil types. Unfortunately, today the mapping of SNRs is yet incomplete at a national scale. Thus a stratification by agricultural region seems to be the most (ope)rational method for future data processing.