Guidance on the assessment of exposure of operators, workers, residents and bystanders in risk assessment of plant protection products

2.3.2. Step 2: use of standardised first‐tier methods of exposure assessment where available

For each risk assessment that is deemed necessary, potential daily exposures should be assessed using standardised methods based on measured data where available. These methods have been defined for the most commonly occurring exposure scenarios, which are specified in terms of:

• ○
  The category of individual exposed – operator, worker, resident or bystander.
• ○
  The type of the PPP – e.g. whether it is formulated as a solid or a liquid.
• ○
  The operations that will be carried out with the PPP and the equipment that will be used – e.g. mixing and loading, application by tractor‐mounted equipment, outdoor/indoor application with handheld application equipment.
• ○
  The intended uses.

In some cases, it may be necessary to combine exposures from two or more activities to obtain a figure for the total potential daily group exposure – e.g. an operator might be exposed both during preparation of the spray solution (mixing and loading) and spraying. Also, individuals may receive different group exposures on the same day (e.g. an operator doing mixing/loading, application and cleaning and also re‐entering a treated field as a worker). However, it is considered sufficient to assume the exposure from the highest exposed group over a single day represents the exposure from two or more activities.

In the case of professional operators and workers, it may be determined that it is necessary to reduce exposure effectively through the use of protective measures. If so, the exposure of these groups should, where possible, be assessed both with and without the proposed protection(s). Measured values from exposure studies should be used, as in the AOEM for example. Otherwise, the multiplying factors by which protection can be assumed to reduce exposures in Table 8 can be used as discussed in Section 2.4.7.1.

2.4.1. Body weights

In all calculations, it should be assumed, as a default, that adults have a body weight (bw) of 60 kg. For children, the body weight of 10 kg for toddlers is used in the calculations (Table 3).

According to the EFSA Guidance on default values (EFSA Scientific Committee, 2012), a body weight of 70 kg should be used as default for the European adult population for consumer dietary risk assessment (over 18 years old). However, when a particular subpopulation is identified as a focus for the risk assessment, actual data for this specific group should be used instead of the proposed default value. Therefore, a default body weight value of 60 kg is proposed in this guidance to be protective for the non‐dietary risk assessment of all adults, including females and teenagers from 14 to 18 years, exposed from professional use of PPPs. The proposed value is in line with the approaches for Biocides (European Commission, 2017a).

Selection of the 10 kg bw value for children is assumed to represent a worst‐case scenario for the risk assessment for children up to 14 years old exposed as residents and bystanders. Children less than a year old, which would be represented by a lower body weight, are not expected to be exposed through entry into treated fields (especially not via the dermal route). Nevertheless, exposure of this age group may occur via hand‐to‐mouth transfer by playing on lawns. However, intensity of contact considering overall exposure will be significantly higher for children > 1 year of age, so that these children are assumed to be the ‘worst‐case’.

2.4.2. Inhalation rates

Where values for potential inhalation exposure are given as concentrations per cubic metre of air, an assumption must be made about the person’s inhalation rate in order to derive an estimate of the inhaled amount and systemic exposure.

For operator and worker, the exposure by inhalation needs to be estimated for a whole working day. Therefore, an average inhalation rate of 1.25 m³/h should be used (European Commission, 2017a) and the daily exposure calculated by multiplying the inhalation rate/h by the working hours per day.

For resident exposure to vapours (longer term exposure), the daily inhalation rate should be taken as shown in Table 4.

As for body weight, the daily inhalation rate of children aged 1 year to less than 3 years of 0.8 m³/day per kg bw was selected to represent the worst‐case scenario across the available scenarios up to 14‐year‐old children and to be protective for other age groups.

For bystander, inhalation exposure could occur predominantly over a shorter period (i.e. typically less than 30 min in duration) and during which activity could be markedly more intense than the daily average; therefore, higher values should be assumed, as shown in Table 5.

As for daily inhalation rate, the hourly inhalation rate of children aged 1 year to less than 3 years of 0.228 m³/hour per kg bw was selected to be the worst‐case scenario across the available scenarios up to 14‐year‐old children and to be protective for other age groups.

2.4.3. Average air concentrations

To estimate 24‐h average concentrations of volatilised pesticides that may be inhaled, different approaches can be used for active substances with vapour pressure (at 20 or 25°C) lower than 10^‐2 Pa.

As first approach, average air concentrations in the 24 h following application are estimated using surrogate field data (California EPA, 1998; Siebers et al., 2003; PSD, 2008; European Commission, 2011) as follows:

• ○
  Substances with low volatility having a vapour pressure of < 5 × 10^–3 Pa, the surrogate default average concentration in air for the 24 h after application is 1 µg/m^3, derived from Siebers et al. (2003).
• ○
  Moderately volatile substances with a vapour pressure between 5 × 10^–3 Pa and 10^–2 Pa, the default average concentration in air for the 24 h after application is 15 µg/m³, derived from California EPA, 1998.

As alternative approach, the saturated vapour concentration (SVC) can be calculated and is assumed to be the worst‐case scenario, as it is not possible for the concentrations in air to exceed the SVC at a given temperature. Such estimates are likely to be very conservative (overprotective) compared to actual concentrations in air (note exposure from breathing in spray, mist or aerosol is assessed separately). This approach is the same as adopted for Biocides (European Commission, 2011).

The SVC should be calculated from the substance’s measured or estimated vapour pressure as follows

where

– SVC = saturated vapour concentration (in mg/m³)

– mw = molecular weight of the active substance (in g/mol)

– vp = vapour pressure (in Pascal)

– R = gas constant = 8.31451 J × mol^‐1 × K^‐1 (physical constant)

– T = temperature = 293 K (assumed room temperature = 20°C)

Resulting in SVC = 0.41 × mw × vp [mg/m³].

Estimates based on this method for substances with very low vapour pressures (i.e. below 10⁻⁵ Pa) are likely to more realistic, although still conservative, than estimates based on the surrogate field data mentioned above.

For active substances with vapour pressures ≥ 10^–2 Pa, since no default value is available, the risk assessments should be based on the SVC approach. If it indicates exposures above the health‐based guidance values, specific measurements of concentrations in air under conditions representative of intended use are required. Alternatively, or possibly in addition, if available, suitably validated dispersion models may be employed to support an exposure assessment.

2.4.4. Hectares treated per day

Table 6 shows default values for area treated per day, in hectares, depending on the type of crop and the application technique. The area treated reflects the technical standard of the equipment used in the original studies underpinning exposure data. In practice, the treated area will depend on the type of equipment used. The assessments proposed for operators, given modern equipment, are also considered to cover the assessment of less modern equipment in correlation with smaller areas treated per day and using smaller amounts of PPPs. The values used for the proposed models should not be adjusted for smaller areas treated with less modern equipment.

For crops not reported in Table 6, further justification has to be provided by the applicant to show the most appropriate scenario to bridge the information to.

The WG agreed that the areas in Table 6 were also applicable to granular formulations.

In the online calculator, the selection of the scenario will automatically select the appropriate treated area per day.

2.4.5. Exposure durations

• ○
  Operator: 8 h.
• ○
  Worker: 2 h (default inspection or irrigation activities); 8 h (other activities, e.g. hand harvesting, thinning, tying, etc.).
• ○
  Resident and bystander: 2 h (dermal, surface deposits), 0.25 h (dermal, entry into treated crops) and 24 h (inhalation from vapour).

2.4.6. Absorption values

Dermal and oral absorption percentages should be taken from the toxicological evaluation.

• ○
  Oral: if less than 80%, the specific value should be entered in the online calculator; if above 80%, the online calculator will automatically consider 100% oral absorption (note that an oral absorption value lower than 80% is also taken into account during the derivation of the toxicological reference value).
• ○
  Dermal: to be determined according to the EFSA Guidance Document on Dermal Absorption, as in force, on the basis of a dermal absorption study or, if no specific study is available, as an appropriate default value. For worker, resident and bystander exposure towards surface deposits and re‐entry into treated crops, the higher of the values for the undiluted product and the in‐use dilution should be taken from the dermal absorption study. The use of higher dermal absorption value is based on the precautionary principle as currently no validated method to measure dermal absorption of dried residue after application of dilutions is available.
• ○
  Inhalation: 100%

A.3.2 Statistical analysis

The technical details of the statistical modelling are clearly and transparently presented in the BfR report (BfR, 2020), and it has been possible to reproduce the results. The use of semi‐parametric quantile regression is consistent with the development of the AOEM model. The argument made for using quantile regression is strong although there are alternatives, such as parametric regression modelling for censored data, which might perform better in some regards but might perform less well in others. Expert judgement, informed by the results of multiple regression modelling, was used to decide which factors appear in each model. Expert judgement is needed because there are no satisfactory automatic statistical methods which can take into account contextual information and background knowledge. However, it should be recognised that choices have been made that might be made differently by others.

The decision to focus on the 75th and 95th percentiles of exposure is consistent with the development of the outdoor AOEM model and with the EFSA Guidance issued in 2014 and the EFSA PPR Panel (2010) Scientific Opinion which stated that ‘for acute risk assessments, exposure estimates should normally be based on 95th centiles of relevant data sets, whereas for longer term risk assessments, the starting point should be a 75th centile’. EFSA PPR Panel (2010) also noted that ‘estimates of exposure from small data sets may be liable to major statistical uncertainty’ and recommended that ‘the exposure value used for risk assessment should be the higher of: (a) the appropriate centile in the relevant data set; and (b) a parametric estimate of the corresponding centile in the theoretical population of measurements from which the dataset was derived’. The procedure specified by EFSA PPR Panel (2010) for the parametric estimate is the upper limit of the relevant prediction interval for an individual exposure, based on using normal distributions to model variability of the logarithm of exposure (for 95th percentile exposure, the relevant prediction interval is the central 90% prediction interval). EFSA (2014) adds a restriction that the exposure estimate should not exceed the maximum of the data. This procedure is principle applicable to the data for knapsack mixing/loading and handheld application in low crops where estimates of percentiles are provided without use of regression modelling. However, it was not possible to apply this procedure systematically to those data because the log‐normal distribution is a poor fit to many of those data sets. There is no obvious alternative to apply to the cases of poor fit, and therefore, this option has not been pursued further.

EFSA PPR Panel (2010) did not consider regression modelling to make exposure estimates dependent on covariates. EFSA (2014) considered the use of regression modelling and recommended the use of quantile regression which would seem to be an appropriate equivalent to the above‐mentioned options (a) and made no statement about a parametric calculation corresponding to (b). A natural approach to (b) would seem to be to use the relevant prediction interval for a conventional multiple linear regression model taking the logarithm of exposure as the response variable. However, the conventional linear regression model would make the inappropriate assumption that the residual variation is homogeneous, and therefore, this approach has not been pursued. An alternative approach to allow for uncertainty in the regression estimates might be to use the upper limit of a confidence interval for the estimate of the relevant percentile of exposure. To implement this would require a decision about the appropriate level of confidence to use and this option has not been pursued.

A difficulty with the use of quantile regression is that it does not require that the model for 95th percentile exposure results in higher exposure estimates than the model for 75th percentile exposure. In most cases, this issue does not arise, but there is one case for which it does and the recommendation in the BfR report is to use the 75th percentile estimate as the 95th percentile estimate when the 75th percentile estimate is higher. This is a pragmatic decision in the circumstances.

Although the total number of replicates is reasonably high, the models rely on data from a moderate number of studies and the statistical modelling does not include components to model study differences. This is effectively a judgement that the studies are sufficiently comparable that differences between measured exposures can be accounted for by the factors included in models and are otherwise random. This is an important judgement given that a number of differences between studies are identified that might have an impact on exposure, e.g. differences in protective clothing. However, inclusion of study effects in the models would increase the difficulty of interpreting of the models for use in risk assessment and there is substantial confounding between study and explanatory factors which would not be easy to address. Cross‐validation was used to validate the modelling and the results for omitting 10% of data from each model and assessing resulting predictions for those data are satisfactory. However, the division of the data into 10 subsets for this purpose is random and a cross‐validation based on omitting entire studies might be expected to show larger differences.

Concerning the use of trolley sprayers, as the trolley equipment does not allow the spraying of the outside rows or other areas where the trolley cannot be driven into (considered as representing 10% of the treated area), these areas have to be sprayed by other means, e.g. lance sprayers. As a consequence, it is considered appropriate first to calculate separate exposure estimates using the model for handheld application with 10% of the total amount of active substance applied and the model for trolley application with 90% of the total amount and then to add these two estimates to obtain the overall exposure estimate for trolley application. To ensure that all cases are covered, it is considered appropriate to always use the dense scenario for the handheld application part of the calculation. For exposure during mixing and loading (before trolley application), the data for tank mixing and loading are considered more robust and more realistic.

A.3.3 Summary results

Results of the BfR Project (BfR, 2015, 2020) are summarised in Tables A.1 and A.2. The results in the BfR report were presented in a transparent way. All exposure information has been published. The statistical evaluation and the corresponding results are comprehensible. Based on the current data availability, both the revised models for mixing/loading and the new models for the application of plant protection products in greenhouses are appropriate for the risk assessment.

D.2.2 Methodologies

For the current update of the guidance issued in 2014 and looking into the Lewis‐DT₅₀ database in more detail, it was agreed that not all the records would be relevant for worker exposure.

More specifically, it had been considered that those records that were captured as ‘IN’ the matrix referred to residue measured in the plant sample (as total residue) and it had been decided to exclude these records.

Thus, it was decided not to filter the ‘Matrix’ parameter but exclude the ‘IN’ records for the ‘IN or ON’ parameter. This filtering had resulted in 746 records out of total 1,048,575, corresponding to 295 publications.

As a first step, the set of these 295 publications was decided to be subject to a more detailed review.

Based on a pilot study, the need for a closer and detailed screening of the records entered in the Lewis and Tzilivakis (2017b) database was identified.

It was decided that for further exploitation of these data, the individual publications should be reviewed based on specific additional criteria, while the existing Lewis‐DT₅₀ database should be restructured in order to include additional information. For example, in many entries of the Lewis‐DT₅₀ database, the same value was recorded as DT₅₀ mean, min and max, while it was not possible to identify those entries derived from a review publication. Furthermore, in specific entries, there was a note that the recorded values had been calculated without providing any relevant information.

Thus, it was decided that the abstracts of all 295 publications identified as ‘ON’ matrix should be screened in order to further confirm their relevance for worker exposure.

Following this step, the need to further confirm the validity of the records captured as ‘IN’ matrix, at least for the matrices related to foliage/leaves/skin surface, was identified. Thus, the abstracts of further 362 publications were screened for their relevance to worker exposure. More specifically, the following were considered:

• Residue to be considered should be only those which are reliably dislodgeable
• Not the total amount of residue but their dissipation rate is crucial
• Only studies with measurements ‘ON’ matrix should be included.

The overall outcome of this prescreening of Lewis and Tzilivakis (2017b) database (1^st selection step) resulted in the need for further assessment of in total 133 publications, 106 of the 279 ‘ON’ screened and 27 of the 362 ‘IN’ prescreened. The detailed selection procedure is described in Annex C.

It is noted that for 16 references from the 295 ‘ON’, neither the publication was available nor the abstract could be retrieved online.

Based on the outcome of this prescreening pilot study, a detailed protocol was developed for further checking the publications identified as relevant (Annex C).

It is noted that for compiling the structure of the new DT₅₀ database, the BROWSE Reports (BROWSE, 2016) for Work Package 2 (Deliverable 2.4 – Work Package 2: Completed worker exposure models for final scenario & Appendices) and the EFSA external scientific report ‘Data collection for the estimation of ecological data (specific focal species, time spent in treated areas collecting food, composition of diet), residue level and residue decline on food items to be used in the risk assessment for birds and mammals’ of Lahr et al. (2018) have been taken into account. In addition, the captured parameters have been chosen considering their potential influence on DT₅₀ results based on elements highlighted also for the higher tier studies (Appendix J).

As a next step, and in order to consider a possible refinement of the default DT₅₀, a data collection was outsourced by EFSA aiming at reviewing the identified published literature from the Lewis and Tzilivakis data set (2017b) by applying the protocol developed by the WG (Annex C).

For the outsourced data collection, 130⁴ publications were checked against eight inclusion/exclusion criteria (Table 2 of the protocol in Annex C):

• Presence of control samples (excluded if no control samples were gathered);
• Number of samples per sampling interval (excluded if less than three);
• Storage conditions (excluded if samples were stored frozen or in dry ice);
• Dislodging solvent (excluded if organic solvents were used);
• Extraction time after sampling (excluded if extraction of DFR was not conducted within 24 h);
• Reporting of the dislodgeable extraction method (excluded if not reported);
• DT50 calculation clearly reported (excluded if not reported);
• Review articles (excluded secondary literature in general)

After the first 20 publications were screened and following consultation with WG members and EFSA, the consideration of one additional screening step was agreed. More specifically, since it was realised that among the 20 studies first checked finally only one concerned the determination of dislodgeable residue, it was decided that before checking for the inclusion/exclusion criteria, the relevance of each publication should be confirmed, i.e. whether the objective of the study was the determination of residues that are reliably dislodgeable and ‘ON’ the matrix. Additional guidance was provided for the data collection regarding the outcome of the prescreening step conducted by EFSA WG, which had been based on reviewing only the abstract for each study.

Finally, 80 out of the 130 checked publications were considered as not relevant to dislodgeable residue on the matrix. Thus, only 50 were the objective of the next step, i.e. the check against the inclusion/exclusion criteria.

Initially, only studies/data fulfilling all eight criteria should have been considered for detailed data extraction in the DT₅₀ database. However, further clarifications were required during the implementation of this check and the need for amending the protocol was identified. More specifically, following a consultation with the WG and EFSA, it was agreed that:

• The criterion ‘Storage other than freeze or with dry ice’ was considered as fulfilled if leaf samples were stored in a cooling box on the way to the lab, or put on ice for the transport, but not frozen.
• If the time until extraction was not mentioned and/or there was no information regarding storage, then the publication should be further considered for data extraction. However, since actually it would be assumed that the samples were extracted following their arrival in the laboratory, the relevant entries should be flagged for further addressing uncertainty issues during the analysis.
• In case there were DFR data available but there was no DT₅₀ calculation, the study was considered acceptable for further review; a DT₅₀ value was calculated by the awarded expert based on the DFR data.
• The lack of control samples, or the absence of reference to control samples was not a criterion for excluding the study for further review and data extraction, but the study was flagged, for further addressing uncertainty issues during the analysis.

Overall, 32 of the 50 publications were considered as relevant for further data extraction in accordance with the protocol.

D.3.1 DT₅₀ values: reported vs. calculated

For data points where DT₅₀ values were reported but also DFR raw data were available, DT₅₀ has been calculated using CAKE. In several instances, the reported and the calculated DT₅₀ values were different; e.g. in study ID 104 the reported DT₅₀ was 30 days, while the calculated DT₅₀ was is 22.5 days. For the majority of the studies, however, good agreement between the reported and the calculated DT₅₀ was noted.

Discrepancies between the DT₅₀ values reported in the publications and calculated in the context of the outsourced data extraction could not be easily explained. However, based upon generally spoken minor differences (Table D.3), no further attempts have been undertaken to explore them. Potential reasons of the noted discrepancies might include the following:

• In the published papers, no detailed explanation on the regression analysis performed was included.
• The calculated DT₅₀ values were based mostly only on averages since raw data were available in limited studies.
• It cannot be excluded that typos were present in the published studies.

Figure D.1 shows the relationship between reported and calculated DT50 values, when both are available (46 data points). The solid line describes the cases where DT50 values are equal and the dashed lines where one is twice the other. While DT50 values from data points considered to be ‘reliable with flagged uncertainties’ tend to have calculated DT50 values greater than reported, the DT50 values from data points concluded ‘reliable with flagged uncertainties*’ tend to have reported DT50 values higher than calculated.

Open in a separate window

Figure D.1

Reported and calculated DT₅₀ values, and differences as regards reliability

D.3.2 General statistics – assessment scenarios

As regards reported vs. calculated DT₅₀ values, different scenarios were initially considered in the evaluations:

• A scenario: statistics for the DT₅₀ values reported from the papers only (137 data points);
• B scenario: statistics for the DT₅₀ values calculated from DFR data using CAKE software (62 data points);
• C scenario: statistics for data set consisting of reported DT₅₀ values and calculated DT₅₀ values where no DT₅₀ values were reported but DFR data available (153 data points);
• D scenario: statistics for data set consisting of reported and calculated DT₅₀ values where calculated DT₅₀ always replaced the reported DT₅₀ if DFR data were available to do so (153 data points);
• E scenario: statistics for all DT₅₀ values, both reported in the public papers and calculated from DFR data (199 data points).

Based on the simple analysis performed (Table D.4), all scenarios revealed comparable statistics.

D.3.3 Studies as regards their reliability

The DT₅₀ data points building the C scenario were evaluated according to the level of reliability assigned, i.e. ‘reliable’, ‘reliable with flagged uncertainties’ and ‘reliable with flagged uncertainties*’ (Table D.5).

The 28 fully reliable data points have lower summary statistics, while the 42 data points graded as ‘reliable with flagged uncertainties’ had higher summary statistics compared to the other two data sets.

The differences in distribution of DT₅₀ values when evaluated for different reliability levels would be statistically significant (Kruskal–Wallis rank sum test, p = 0.0061) if data were treated as randomly sampled and unstructured (Figure D.1). However, since the variables, such as active substance and crop, are not the same in the three reliability groups, the differences in distribution could be due to a combination of other factors and not due to reliability rating.

Figure D.2 hows the distribution of DT₅₀ values in histogram and highlights the reliability status of data points. As regards the absence of higher DT₅₀ values in the category of ‘reliable’ studies, it cannot be excluded that this is partially due to the lower number of data here (28 data points) (Figure D.3)

Open in a separate window

Figure D.2

Distribution of DT₅₀ values as regards reliability of data points

Open in a separate window

Figure D.3

Histogram of DT₅₀ values as regards reliability of data points

No calculated DT₅₀ values are available for data points concluded as ‘reliable with flagged uncertainties*’. Although there is no clear picture as regards differences between the three reliability groups, DT₅₀ of 30 days is a high value regardless which group of data is considered for the calculation, i.e. reliable or reliable with flagged restrictions (with or without asterisk) (Figure D.4).

Open in a separate window

Figure D.4

Histogram of DT₅₀ values as regards reliability of data points

D.3.4 Analysis of potential factors influencing DT₅₀ values

The database containing reliable studies contains data for 32 different active substances, and 24 different crops. Among the 32 active substances, only seven are still approved in the EU (status: February 2021), i.e. abamectin, azadirachtin, bupirimate, captan, emamectin, esfenvalerate, ethofumesate (Regulation (EU) 1107/2009). From the 25 active substances in the database which are not approved in the EU, only for endosulfan one the reasons for non‐approval was ‘persistence’. For the other 24 substances either different justifications for non‐approval were listed or no reasoning could be captured in the pesticide database (https://ec.europa.eu/food/plant/pesticides/eu‐pesticides‐db_en) without deeper investigations. As regards the 28 fully reliable data points, they cover only six active substances currently approved in the EU.

The distribution of DT₅₀ values reveals statistically significant differences between different crops (Kruskal–Wallis test, p = 4 × 10⁻¹²) and/or between different active substances (Kruskal–Wallis test, p = 5 × 10⁻¹¹), but these two factors are heavily confounded, and it is not possible to distinguish which of these might be driving variation in the DT₅₀. The same applies to whether the study was in the field or indoor, the mode of action and the matrix (leaves, fruits) since these covariates are also quite unbalanced: the vast majority of data are from the field, are for insecticides and are on leaves. Based upon available data, their amount and quality, it is not possible to decide which relationships are real and which are due to confounding. Some general evaluations are included below.

D.3.5 Indoor–outdoor

Among the 153 data points, only 11 covered indoor applications. None of the active substances was tested both outdoor and indoor. Based on the limited number of data points for indoor applications, no evaluation could be further done to reliably analyse the influence of indoor conditions, but 75th and 95th percentiles DT₅₀ values were remarkably higher than for field conditions.

Among the 142 data points from outdoor applications, 122 concerned dislodgeable residue on leaves while 20 concerned dislodgeable residue on fruits. No data points related to dislodgeable residue on fruits for indoor application are available. Although very limited data points were available for fruits, the statistics was not different from the one done on DT₅₀ values on leaves (Table D.6).

D.3.6 Crops and matrix (leaves, fruits)

In most of the studies, the crops were in a mature stage, so it can be concluded that no ‘diluting’ effects on DT₅₀ values occurred due to the plant growth after the treatment period. Only for study ID 115, captan was applied through the growing period.

Twelve active substances were tested at least on two crops. The majority of these data points (i.e. 94) was obtained from dislodgeable residue on leaves, while only seven data points were from fruit surface. Fruit surface DT₅₀ values were not available for more than one crop for each active substance. The limited amount of data points available do not allow the conclusion if residue on fruit surfaces was higher than on leaves, as regards the same active substance.

The DT₅₀ values measured for the same active substance from different crops on the same matrix (leaves) were very well comparable between the crops, i.e. they were of same magnitude (Table D.7). Only in case of azinphos‐methyl, where one value was available for peach leaves, whilst nine values were available for pear leaves, the 9 DT₅₀ values for pear leaves spread from 3 to 24 days, showing high variability in dissipation rate in this case.

It is further noted that the 153 DT₅₀ values cannot be considered as representative for all crops; only data for 23 crops were obtained as reliable. While there are 40 data points for grape/vine and 39 data point for citrus fruits (grapefruit, lemon, orange), there are only one to three data points for crops such as alfalfa, apple, cucumber, cauliflower, green onion, tomato and broccoli.

D.3.7 Active substance – crop(s)

All crops were evaluated individually, only citrus fruits (oranges, grapefruits, lemons) were combined. The data indicate that dissipation rate is probably less driven by the crop properties than by the active substance properties. However, too few data are available to further substantiate this assumption. In Table D.9 only data points where more than one active–substance–crop combination was available are included.

D.3.8 Seasons – geographical area

Potential influence of season or geographical area was investigated only for field applications (Table D.10).

For azinphos‐methyl individual DT₅₀ values ranged from 3.7 to 30 days, showing a wide range of dissipation rates for pear leaves. Since data for pears were collected only from one location from May to June, no influence of seasons nor of geographical areas can be attributed to this difference in dissipation values.

The geographical areas captured in this analysis, all from North America, are considered relevant for European conditions. There was only one study conducted in India (equatorial climate according to Köppen–Geiger criteria), which was not considered representative for European climate zone.

D.3.9 Adjuvants; physical state of the formulations

All DT₅₀ values (except two) were determined from spray application of products. Therefore, the impact of different types of application (e.g. spraying vs. dusting, etc.) on the DT₅₀ values cannot be investigated. Effect of adjuvants cannot be determined neither since only the information is very scarce (adjuvant used only in three studies with three different active substances). Only for emamectin, DT₅₀ values were obtained with and without the use of adjuvants, revealing values of 0.6 and 0.4 d, respectively. The database lacks sufficient number of representative values to conclude on the possible effects of adjuvants on DT₅₀ values.

D.3.10 Meteorological conditions

The most frequent information on meteorological conditions was on temperature and rain. Rain occurred either long after the last application when residue levels were low or before the last treatment. Heavy rain was not reported but, generally, light rain episodes were described. The DT₅₀ values, obtained in trials where rain occurred, were considered not to be severely affected by this variable.

Information on wind conditions was scarcely reported. Wind measurements were reported only in two studies and only in one the wind speed was reported. Wind data, when available, were not discussed by the study authors.

Temperature mean data are available for only four active substances as reported in Table D.11 below.

D.3.11 Analytical methods

Information regarding the dislodgeable residue extraction methods (including the non‐organic solvents as aqueous dilutions) was available for all the studies considered during the data extraction phase. On the other hand, methods for quantification of DFR and LOQ/LOD in reliable studies were available only for limited numbers of data points (21 LOQs and 97 LODs).

As regards the correction of field results for recovery, this information was available for 101 DT₅₀ values out of 153 composing the database. For the remaining 52 DT₅₀ values (34%), no information in this regard was retrieved.

F.4.1 Vegetables

The EFSA TC is 2,500 cm²/h, and this applies to harvesting both outdoor and protected (greenhouse) crops. The EPA greenhouse vegetable TC are lower. The EPA hairy‐leaf and smooth‐leaf (except sweet corn) harvest activities are also lower. EPA TC for harvesting low height waxy‐leaf crops are also within the EFSA value.

The EFSA TC appears not to be protective for harvesting sweet corn in cases where there is potential for intense contact. The observed 75th centile was over 920% of the EFSA value. Similarly, the EPA TC for all activities in medium height waxy‐leaf crops frequently exceed the EFSA TC and the 75th centile was about 200% of the EFSA value.

F.4.2 Tree fruits

The EFSA TC is 4,500 cm²/h. Most of the EPA TC observed in orchard crops lie below this value. The thinning activity gave the highest TC, and the observed 75th centile is slightly higher than the EFSA value.

F.4.3 Grapes

The EFSA TC is 10,100 cm²/h. This was established in part by reference to information from the USA. The EPA data for harvesting cane berries are clearly lower than the EFSA value. The data for harvesting raisin grapes also exhibit a 75th centile lower than the EFSA value.

The 75th centile for harvesting juice/wine grapes is about 20% higher than the EFSA value. Although this difference is not considered to be significant. The EPA have also established a cluster for more intense contact activity (cane turning) that has a 75th centile 240% of the EFSA value.

F.4.4 Hops

The US data do not include exposure measurements to support TCs for work activities in hops. However, the US EPA and the Canadian authorities both originally used the TC derived for cane turning in grapes (considered to be the most intense contact cluster in grapes) for hop harvesting. Then, in 2015, the US EPA, based on information regarding the high level of mechanisation in the USA, changed their position to accept a much lower TC, representative of hand and forearm exposure only, for mechanically assisted hop harvesting, and also concluded that no TC was required for fully mechanical hop harvesting (US EPA, 2017). However, Health Canada reviewed harvesting practices in Canada and found less mechanisation and potential for high exposure activities during cutting bines such as loading the bines into the truck/trailer, which included body exposure, loading the bines on the stripping table and stripping by hand. As such activities could involve significant contact between foliage and workers, Health Canada did not revise the TC for mechanically assisted hop harvesting. They did accept in some situations hop harvesting could be completely mechanical.

Information on European hop harvest practices shows similar work practices to those reported in Canada, with a potential for body and hands foliage contact when cutting and loading bines into the trailer, followed by potentially hand and forearm contact when stringing the bines up prior to mechanical picking. As such, mechanically assisted hop harvesting is not considered to be a no or very limited worker exposure scenario. In the absence of specific data measured on hops, the current EFSA TC for grape harvesting is assumed to be a suitable surrogate for harvesting hops.

F.4.5 Strawberry

The EFSA TC is 3,000 cm²/h. Strawberries were one of the crops where exposure was monitored in the smooth‐leaf field crop harvest cluster. The observed 75th centile was less than half the EFSA value.

F.4.6 Ornamentals

The EFSA TC is 5,000 cm²/h. The EPA cluster for floriculture hand harvesting has an observed 75th centile that is marginally below this value.

F.4.7 Golf course, turf or other sports lawns

The EFSA TC is 2,500 cm²/h. The EPA have established two TC clusters for turf. The cluster for golf course maintenance exhibits a very wide spread in the data (the EPA considered there was insufficient data to create subcategories showing less variation and the EPA also noted that daily work routines typically involve mixed maintenance activities, so it may be inappropriate to assume only a limited range of actions). The observed 75th centile is lower than the EFSA TC. The second cluster is for turf farming.

The data available for turf (sod) harvesting show less variation than the golf‐course data, and they appear to be typically higher. The observed 75th centile TC value is 350% times the EFSA TC.