**3. Key farming practices**

bles. One explanation for the observed effect is that CLD is trapped in the microstructure of the allophane clay, thus reducing its transfer from the soil to the plant. The influence of soil allophane content on CLD retention is the signature of the peculiar microstructure of the al‐ lophane aggregate. The spongy structure influences transport inside the allophane aggre‐ gates. The SAXS data made it possible for us to propose a mechanism for the retention of pesticides in allophanic soils and also for the limited release of CLD to crops and water re‐ sources [1, 19]. At the scale of the allophane, accessibility is difficult because of the fractal structure and small pore size of allophane clay. CLD transfers within the soil depend on hy‐ draulic conductivity (*K*) and diffusion processes (*Di*) in the porous microstructure. The frac‐ tal structure allows an approximation of *K* and *Di* at the aggregates scale *l* [26] through the

( ) ( ( ) ) ( ( ) )( ) <sup>3</sup> <sup>3</sup> <sup>3</sup> 3 2 3 2 <sup>1</sup> and 1 1 232 / [ ) ]. *Df Df Df Df K l l a l Di l a l a la* - - - - ¥ é é - <sup>ù</sup> <sup>ù</sup> ¥ ë ë - <sup>û</sup> - <sup>û</sup> - (1)

We calculated the transport properties (hydraulic conductivity and diffusion, inside the allo‐ phane fractal aggregate, i.e. between 3 and 60 nm (Table 2). Figure 5 shows changes in *K* and *Di* normalized to *K* and *Di* at *l* = 60 nm. Hydraulic conductivity decreased by 4 orders of magnitude and *Di* decreased by 20 orders of magnitude when *l* decreased from 60 to 4 nm.

The very low calculated *Di* and *K* suggest that CLD is trapped in the porosity. The trapping mechanism is favored by the large size of the fractal labyrinth. The higher the allophane content, the bigger the labyrinth and the higher the retention. Like nanoporous materials [28], the paradox of allophane clay is that it has large porosity but poor transport properties. In these fractal structures, possible reactions with chemical or biological species that could extract the pesticides are thus hindered; the pesticide remains trapped inside allophane clay

following equations:

620 Environmental Risk Assessment of Soil Contamination

**Figure 5.** Relative K (○) and Di (▲) versus the scale length, l (nm)

and cannot be extracted.

Soil type determines the potential for CLD sequestration but, in the analysis of risk and in the diagnosis of soil pollution, farming practices are key factors that determine the level of the pollutant stock, its potential availability, and its potential transfer to different environ‐ mental compartments [1, 10, 33]. In this section, we examine the effect of three main types of practices characteristic of past and present farming practices. The first is the level of intensi‐ fication of the cropping system, which, in the case of CLD, determines the initial input [1, 10, 33]. The second is soil tillage, which determines the depth (and hence the volume) of soil affected by the pollution [1, 10, 33]. The third is organic matter amendment, which affects the availability of CLD [34].

#### **3.1. Level of intensification of the cropping system**

It is a truism to say that soils are polluted because they have received pollutants. Farmers use a wide range of strategies based on pesticides to protect their crops. These strategies mainly depend on the intensification of the cropping system [35-37]. Intensive systems pro‐ duce a higher cash flow and more profit than small diversified systems [38, 39] and farmers tend to minimize the risk of yield loss by intensifying chemical pest control [10]. These farm‐ ers have to deal with higher pest pressure than systems that include crop rotation and diver‐ sification, and consequently require more frequent treatments [40, 41]. In agro-industrial banana plantations in the French West Indies, agronomic and economic conditions led to the intensive use of agrochemicals [39] i.e. frequent applications and/or high doses of the persis‐ tent molecule CLD over large areas. In 2013, twenty years after the treatments ended, the ubiquitous presence of the CLD in different environmental compartments raises the ques‐ tion of the impact of these past applications on soil contamination and its true extent.

To answer this question, the link between CLD supply and soil contamination was first ex‐ amined using a simple CLD leaching model called WISORCH [1]. This model predicts soil CLD content based on the history of CLD applications under different farming systems, dif‐ ferent soil types, and different average annual rainfalls. The model accounts for andosol, ni‐ tisol and ferralsol through their main characteristics, notably the soil-water partition coefficient relative to organic C content (*Koc* in m3 kg-1), soil organic carbon content, depth of tillage and soil bulk density.

Simulation results first showed that the schedule of CLD applications, i.e. CLD loads, had the most impact. Long after the application of CLD, the effect of CLD loads on soil CLD con‐ tent meant that soil decontamination was extremely slow. By exploring different assump‐ tions, the WISORCH model provided two main explanations for the slowness of decontamination: 1) the absence of degradation; 2) only lixiviation by percolation water can slowly reduce soil contamination. These assumptions are consistent with the lack of evi‐ dence for natural degradation of CLD reported in the literature [9]; given the very low vola‐ tility of CLD, water is the only vector of CLD dispersion.

In addition to the main effect of CLD applications, WISORCH simulations identified tillage depth as the second factor that influences soil CLD content. This factor depends on cropping systems and is discussed below. Here we simply note that physical factors like soil gravi‐ metric carbon content, bulk density and average annual rainfall have less impact. Conse‐ quently, human activity was the first determinant of soil contamination.

The WISORCH model explained why large areas are still contaminated even though treat‐ ments were halted long ago. As the model first identified the schedule of pesticide applica‐ tions, this suggested that tracing the history of these applications would help assess the level and distribution of soil contamination at a regional scale. A historical analysis was per‐ formed in Guadeloupe using maps of banana plantations at different dates. The results re‐ vealed three gradients of land use for banana in 1 145 plots (1 376 ha) that were analyzed for the presence of CLD [10].

Overall, plot contamination increased with the duration of land use in andosols and ferral‐ sols. Table 3 shows that on average, plots with short term banana land use were less conta‐ minated than those with medium and long term use. The lack of a significant difference between medium and long term banana land use in CLD stocks and concentrations was con‐ sistent with the widespread use of CLD in the 1980s and the 1990s, but this was not the case in the 1970s, when CLD was used frequently but not systematically. The effect of banana land use duration was significant for andosols and ferralsols but not for nitisols. This differ‐ ence in behavior between soils could be explained by their ability to retain CLD. In this case, unlike andosols and ferralsols, the lower retention capacity of nitisols could mask the impact of variations in CLD inputs with respect to the length of time the land was used for banana cultivation.


ubiquitous presence of the CLD in different environmental compartments raises the ques‐

To answer this question, the link between CLD supply and soil contamination was first ex‐ amined using a simple CLD leaching model called WISORCH [1]. This model predicts soil CLD content based on the history of CLD applications under different farming systems, dif‐ ferent soil types, and different average annual rainfalls. The model accounts for andosol, ni‐ tisol and ferralsol through their main characteristics, notably the soil-water partition

Simulation results first showed that the schedule of CLD applications, i.e. CLD loads, had the most impact. Long after the application of CLD, the effect of CLD loads on soil CLD con‐ tent meant that soil decontamination was extremely slow. By exploring different assump‐ tions, the WISORCH model provided two main explanations for the slowness of decontamination: 1) the absence of degradation; 2) only lixiviation by percolation water can slowly reduce soil contamination. These assumptions are consistent with the lack of evi‐ dence for natural degradation of CLD reported in the literature [9]; given the very low vola‐

In addition to the main effect of CLD applications, WISORCH simulations identified tillage depth as the second factor that influences soil CLD content. This factor depends on cropping systems and is discussed below. Here we simply note that physical factors like soil gravi‐ metric carbon content, bulk density and average annual rainfall have less impact. Conse‐

The WISORCH model explained why large areas are still contaminated even though treat‐ ments were halted long ago. As the model first identified the schedule of pesticide applica‐ tions, this suggested that tracing the history of these applications would help assess the level and distribution of soil contamination at a regional scale. A historical analysis was per‐ formed in Guadeloupe using maps of banana plantations at different dates. The results re‐ vealed three gradients of land use for banana in 1 145 plots (1 376 ha) that were analyzed for

Overall, plot contamination increased with the duration of land use in andosols and ferral‐ sols. Table 3 shows that on average, plots with short term banana land use were less conta‐ minated than those with medium and long term use. The lack of a significant difference between medium and long term banana land use in CLD stocks and concentrations was con‐ sistent with the widespread use of CLD in the 1980s and the 1990s, but this was not the case in the 1970s, when CLD was used frequently but not systematically. The effect of banana land use duration was significant for andosols and ferralsols but not for nitisols. This differ‐ ence in behavior between soils could be explained by their ability to retain CLD. In this case, unlike andosols and ferralsols, the lower retention capacity of nitisols could mask the impact of variations in CLD inputs with respect to the length of time the land was used for banana

kg-1), soil organic carbon content, depth of

tion of the impact of these past applications on soil contamination and its true extent.

coefficient relative to organic C content (*Koc* in m3

tility of CLD, water is the only vector of CLD dispersion.

quently, human activity was the first determinant of soil contamination.

tillage and soil bulk density.

622 Environmental Risk Assessment of Soil Contamination

the presence of CLD [10].

cultivation.

**Table 3.** Effect of soil type and banana land use duration (short, medium and long term) on the log of mean soil CLD concentration (corresponding values in mg kg-1 in brackets). Differences in means were assessed at two levels: globally, for the "mean soil" column and "mean duration" rows (in italics); for each soil, i.e. each row (in normal font). For each case, the letters a, b and c indicate significant differences between factor levels at p <0.05 (Kruskal-Wallis test). [10]

The soil sorption capacity was first assessed by inverting the WISORCH model [1] to deter‐ mine *Koc*. Results showed that andosols had a higher sorption capacity (*Koc* of 12-25) than fer‐ ralsols (*Koc* of 7.5-12) and nitisols (*Koc* of 2-3).

This is consistent with our previous results concerning CLD sorption. Indeed, these differen‐ ces in *Koc* do not stem from the chemical composition of the clays, since the allophane in an‐ dosols does not significantly differ from phyllosilicate clays (nitisols and ferralsols) [42]. The higher apparent *Koc* in andosols could be the result of the allophane microstructure, leading to lower CLD availability.

Moreover, large farms were more contaminated than small farms [10]. We selected farms with more than 10 plots to assess the intra-farm variation in CLD concentrations. Figure 6 shows that inter-farm variation in CLD concentration was far higher than the intra-farm var‐ iation whatever the soil type, meaning the farm factor was decisive in explaining the distri‐ bution of CLD concentration. Regardless of the type of soil, the treatment strategy used on a given farm was an essential component of soil contamination.

Concerning the type of soil and other physical factors of CLD retention in soils, at regional scale, the effect of the type of soil was clear. Analysis of CLD concentrations (Table 2) re‐ vealed significant differences between soils (p<0.05 Kruskal-Wallis test) showing that ando‐ sols were the most contaminated. Values of CLD concentrations in andosols were 2.3-fold higher than in nitisols, and 3.6-fold higher than in ferralsols. Organic carbon (OC) content was only calculated for andosols to avoid possible interactions with other physical variables (notably, allophane). Results showed that andosols with high OC content tended to retain CLD better. However, the relationship was weak. A probable explanation is that OC had less impact due to the variability of inputs. Finally, although carbon is considered to deter‐

**Figure 6.** Distribution of CLD concentration on farms located on andosols and nitisols with more than 10 banana plots. Each box represents a farm. The bottoms and tops of the boxes represent the 25th and 75th percentile; the band inside the box is the median; the whiskers extend to the most extreme data point, which is no more than 1.5 times the interquartile range of the box [10].

mine sorption of CLD, at the regional scale, organic carbon was not a prime factor in ex‐ plaining variations in CLD concentrations.

To summarize, findings at the regional scale were consistent with the findings of the WI‐ SORCH model at the plot scale. Farming systems mainly explained soil CLD contamination, and physical factors like soil carbon content had less influence. Concerning the type of soil, although the findings were the result of observations at different scales - region, plot, micro‐ structure – they all highlighted the specific sorption capacity of andosols for CLD.

These important observations made it possible to draw maps of areas with a risk of CLD contamination in Guadeloupe and Martinique based on soil type and on an historical analy‐ sis of the supply of CLD [43, 44]. They also identified a pollution system characterized by remarkable inertia comprising (i) a persistent molecule – CLD; (ii) intensive large scale ap‐ plications; (iii) andosols with high organic matter content and an allophanic microstructure that can trap pollutants. This inertia was assessed by simulating changes in soil CLD con‐ tamination using the WISORCH model. Figure 7 shows that one hundred years will be needed to clean up nitisols, and six hundred years to clean up andosols (Cabidoche et al., 2009).

**Figure 7.** Simulation of changes in soil CLD contamination in Guadeloupe

Figure 7 highlights the need to use different temporal scales to assess pollution: a long term scale when soil retention properties are probably the main explanation for the spatial varia‐ bility of soil contamination; short and medium term scales when applications of the pollu‐ tant explain most spatial variability. From a management point of view, authorities are thus justified in focusing on reducing pollutant loads (the frequency of application and the quan‐ tities of pollutant used) on cropped areas, even in the case of less persistent molecules. In the case of CLD, the persistence of pollution calls for further research on soil decontamination. However, in the meantime, different agricultural practices can help manage the risk of con‐ tamination, this being the case of soil tillage and organic matter amendment.

#### **3.2. Tillage practices**

mine sorption of CLD, at the regional scale, organic carbon was not a prime factor in ex‐

**Figure 6.** Distribution of CLD concentration on farms located on andosols and nitisols with more than 10 banana plots. Each box represents a farm. The bottoms and tops of the boxes represent the 25th and 75th percentile; the band inside the box is the median; the whiskers extend to the most extreme data point, which is no more than 1.5

To summarize, findings at the regional scale were consistent with the findings of the WI‐ SORCH model at the plot scale. Farming systems mainly explained soil CLD contamination, and physical factors like soil carbon content had less influence. Concerning the type of soil, although the findings were the result of observations at different scales - region, plot, micro‐

These important observations made it possible to draw maps of areas with a risk of CLD contamination in Guadeloupe and Martinique based on soil type and on an historical analy‐ sis of the supply of CLD [43, 44]. They also identified a pollution system characterized by remarkable inertia comprising (i) a persistent molecule – CLD; (ii) intensive large scale ap‐

structure – they all highlighted the specific sorption capacity of andosols for CLD.

plaining variations in CLD concentrations.

times the interquartile range of the box [10].

624 Environmental Risk Assessment of Soil Contamination

A second step of the diagnosis is the analyses of soil tillage The heterogeneity of CLD con‐ tent at field scale and the effect of tillage were investigated [33].

In the French West Indies, different tillage practices are used on banana plantations. These range from no tillage, especially on sloping plots, to regular deep tillage to a depth of 60 cm or more, every four years [10, 45]. The mode of application of the pesticide (powder spread on the ground around the foot of the banana tree), and the semi-perennial arrangement of trees [46] account for the high heterogeneity observed at field scale. Indeed in our study, in the same plot, CLD contamination of the upper soil layer (0-30 cm) could range from 0.2 to 2.7 mg kg-1 and, in plots of less than 1 ha, from 2.9 to 17.6 mg kg-1.

**Figure 8.** Ratio of CLD contents in the 0-30 cm and 30-60 cm soil layers as a function of the depth of tillage, adapted from [33].

Figure 8 shows that, whatever the soil type, the upper soil layer (0-30 cm) was generally more contaminated than the lower layer (30-60 cm). With no tillage, the 0-30 cm layer was four times more contaminated than the 30-60 cm layer (21 mg kg-1 and 5 mg kg-1). We also showed that tillage had a significant effect (P <0.0001) on the horizontal distribution of the contaminant, leading to pesticide dilution in the soil profile. With deep tillage (60 cm and deeper), whatever the type of soil and mean CLD content, CLD content was similar at the two sampling depths, with mean values of 11 vs. 12 mg kg-1, 12 vs. 11 mg kg-1 and 2 vs. 2 mg kg-1. Thus CLD content was homogenized in the 0-60 cm layer. In plots where tillage was shallower, the upper layer was still significantly more contaminated than the lower layer al‐ though the proportion depended on the tillage depth. This result is in accordance with re‐ sults observed for DDT [47].

Likewise, tillage tended to reduce CLD horizontal heterogeneity. For this reason, when sam‐ pling soil, it is important to take such heterogeneity into account at intra-field scale for accu‐ rate assessment of CLD content. For sampling, it is recommended to subdivide all plots of more than 2 ha and, more generally, to use an appropriate sampling procedure that takes into account landscape (slope and resulting erosion), field history (tillage, cropping system, former inter-row distance, etc.) but also the reasons for sampling: cropping system manage‐ ment or analysis of overall risk.

Tillage is of major concern because it modifies the vertical distribution of the pollutant with‐ in the soil profile, and hence the volume of soil that is contaminated and the level of contam‐ ination. The risk of the pollutant being transferred to the crop depends on the type of soil and soil CLD content [19, 26]. Thus, in some cases (mainly low soil CLD content), this will determine the range of crops that can be cultivated while respecting regulatory thresholds, in particular the maximum residue limit of 20 µg kg-1 fresh matter for food products [23].

It is also important to keep in mind that tillage practices may negatively affect CLD seques‐ tration and distribution. Indeed tillage during the dry season can cause surface desiccation and reduce pore volume, which will irreversibly alter the micro-structure of allophane [45, 48], thus possibly modifying soil CLD sequestration potential. Moreover, inappropriate till‐ age practices increase the risk of erosion, driving the top soil layer downslope [49] thereby modifying CLD distribution at plot scale, with higher CLD content at the bottom of the slope than at the top [33].

In conclusion, analyzing past and present farming practices provides insight into CLD con‐ tent and distribution at intra-field scale. These practices can also affect CLD availability by modifying soil retention properties.

## **3.3. Pesticide sequestration by compost addition**

In the French West Indies, different tillage practices are used on banana plantations. These range from no tillage, especially on sloping plots, to regular deep tillage to a depth of 60 cm or more, every four years [10, 45]. The mode of application of the pesticide (powder spread on the ground around the foot of the banana tree), and the semi-perennial arrangement of trees [46] account for the high heterogeneity observed at field scale. Indeed in our study, in the same plot, CLD contamination of the upper soil layer (0-30 cm) could range from 0.2 to

**Figure 8.** Ratio of CLD contents in the 0-30 cm and 30-60 cm soil layers as a function of the depth of tillage, adapted

Figure 8 shows that, whatever the soil type, the upper soil layer (0-30 cm) was generally more contaminated than the lower layer (30-60 cm). With no tillage, the 0-30 cm layer was four times more contaminated than the 30-60 cm layer (21 mg kg-1 and 5 mg kg-1). We also showed that tillage had a significant effect (P <0.0001) on the horizontal distribution of the contaminant, leading to pesticide dilution in the soil profile. With deep tillage (60 cm and deeper), whatever the type of soil and mean CLD content, CLD content was similar at the two sampling depths, with mean values of 11 vs. 12 mg kg-1, 12 vs. 11 mg kg-1 and 2 vs. 2 mg kg-1. Thus CLD content was homogenized in the 0-60 cm layer. In plots where tillage was shallower, the upper layer was still significantly more contaminated than the lower layer al‐ though the proportion depended on the tillage depth. This result is in accordance with re‐

Likewise, tillage tended to reduce CLD horizontal heterogeneity. For this reason, when sam‐ pling soil, it is important to take such heterogeneity into account at intra-field scale for accu‐ rate assessment of CLD content. For sampling, it is recommended to subdivide all plots of more than 2 ha and, more generally, to use an appropriate sampling procedure that takes into account landscape (slope and resulting erosion), field history (tillage, cropping system, former inter-row distance, etc.) but also the reasons for sampling: cropping system manage‐

2.7 mg kg-1 and, in plots of less than 1 ha, from 2.9 to 17.6 mg kg-1.

626 Environmental Risk Assessment of Soil Contamination

from [33].

sults observed for DDT [47].

ment or analysis of overall risk.

The use of soil organic matter to control the environmental mobility and fate of pesticides has already been reported in the literature [50-53]. Here, we propose an alternative strategy which is quite the opposite of total soil decontamination: CLD sequestration enhanced by soil organic amendment. Because CLD is tightly trapped in the soil, an alternative solution to decontamination may be to further increase its sequestration in the soil thereby reducing pesticide diffusion into the environment. This could be a way to reduce further release of CLD from contaminated soils towards other environmental compartments until efficient re‐ mediation techniques become available. We now examine the hypothesis that adding organ‐ ic matter to contaminated soils improves their CLD sequestration ability with the objective not of removing the pesticide from the soil but rather of controlling its release into the envi‐ ronment.

As detailed above, combined with high organic matter content, the microstructure and the large specific surface area of clay favor the accumulation of pollutants in the soil. In allo‐ phanic soil, the high CLD content is the result of the combination of CLD's high affinity for the soil organic content and the poor accessibility of CLD into the mesopore structure.

With the aim of preventing consumer exposure, we tested the incorporation of compost in soils as a possible way to reduce plant contamination [34], based on the hypothesis that add‐ ing organic matter would improve CLD trapping and thus reduce its bioavailability for crops. We characterized the transfer of CLD from soil to radish, a crop belonging to roots and tubers, a CLD sensitive group. Two months after incorporation of the compost, the con‐ tamination of the different plants organs was 3, 15 and 5 times lower in small roots, tubers and leaves, respectively than without added compost (Figure 9.).

**Figure 9.** CLD transfer in radish organs (leaves, tubers and fine roots) with added compost (in grey) and without (in white).

These experiments also showed that adding compost closed the microstructure of allophane clays, thus favouring CLD retention in allophanic soils. Adding compost altered the porosi‐ ty of the allophane clay in the size range 10 to 60 nm, while the intensity of this effect varied with the allophane content. We suggest that these pore changes are the consequence of ca‐ pillary stress and of the low mechanical properties of the fractal structure [54].

Organic matter (OM) amendments aim to modify chemical conditions within the soil pro‐ file. As mentioned above, the OM soil status influences the availability of the pollutant, a factor of primary importance for both pollutant transfer and degradation [55]. At field scale, OM amendments modify the potential sequestration of persistent organic pollutants in the soil by enhancing the soil's sorption capacity for CLD [56]. This sorption capacity depends on the quantity of OM supplied, the type of OM (stable vs. labile) and the frequency of the application. This practice needs to be studied over time, as OM degrades and could modify the OM – pollutant relationships.

Finally, all these practices depend on farm strategies. When the aim is to modify these prac‐ tices, both the scale and the type of farm need to be taken into account. In the case of CLD pollution, analyses revealed a strong "farm effect" [10]. A typology should be built includ‐ ing the farms' overall strategy and objectives, the types of crops grown, practices (more or less intensive) and the farms' specific field orientation (how often the land is used for each crop).
