Assessment of Spatiotemporal Distribution of Herbicides in European Agricultural Land Using Agri-Environmental Indices

Abstract

Short-term estimates are not suitable for monitoring and comparing fluctuating pesticide use in EU agricultural land. A discriminative and comparable (HI) herbicide index was evaluated to elucidate herbicide use in the 21st century. The HI was 0.66 kg of active substances per hectare of conventional agricultural land across the EU. However, the HI varied between the 27 EU Member States. The highest mean values of HI were observed in Belgium, the Netherlands, Cyprus, Germany, France, and Denmark, with the lowest in Romania, Greece, Bulgaria, and Latvia. The results showed that the distribution of the HI variable was independent of the geographical location of each country, such as from North to South or from West to East in the EU. It seems that country-level agri-environmental parameters ultimately influenced the herbicide use. To assess the causes of this variability, 31 agri-environmental parameters (formatted into indices to be comparable) were investigated, emphasizing the structural characteristics of the agricultural sector in each EU Member State. Using only the significant independent variables (13 out of 31), linear discriminant analysis (LDA) was applied to explore the differentiation potential of EU₂₇ by creating a discrimination model. The assessment of each one variable in the HI could contribute to the reduction in environmental impacts and the faultless implementation of the European agricultural policy in the near future.

1. Introduction

Herbicides and their use are an important aspect of modern weed management techniques among many methods for solving most weed problems. Although in recent decades the use of chemical herbicides has also contributed to the increase in European food production, herbicides are no longer considered friendly, but rather a threat to humans due to their irresponsible use. Thus, intensive use of herbicides causes risks to human health, pollution of water and soil resources, loss of biodiversity, and a headache to farmers due to the loss of productivity and economic losses caused by herbicide-resistant weed populations [1]. Herbicides account for more than 30% of all pesticides used in the EU [2]. In this context, each of the 27 EU Member States has drawn up a national action plan for sustainable weed management, applying multiple control tactics and alternative methods to combat weed infestations, including integrated weed management (IWM) as well as biological methods for weed control in agricultural land and urban or other areas [3].

Data obtained for the period 2000–2021 by the FAO (Food and Agriculture Organization) showed that the EU agricultural land, where the herbicides were widely used, was covered by 162,907,753 hectares, reduced by 10.8%, compared to 2000. Most of this EU agricultural land was covered by cropland (68.7%), presenting a decrease of 9.4% for the above examined period [4]. Despite the reduction in agricultural land in the EU (from 2000 to 2021) as well as the implementation of the above actions aimed at reducing the use of herbicides at the national level, sales of herbicides increased by approximately 11.9% from 2000 to 2021 in the 27 EU Member States [5]. Antier et al. [6] reported that the most widespread used herbicide in the world with the active substance glyphosate increased its total sales in the EU₂₈ countries by 6% from 2013 to 2017.

Herbicide use on agricultural land is influenced by numerous agro-environmental parameters such as agricultural land use (e.g., crop land versus pastureland), farming systems (e.g., organic versus conventional), land cover (e.g., horticultural versus industrial crops), growing season, herbicide chemical families, herbicide-resistant weeds, farming structure, climate conditions, herbicide cost, and EU environmental legislation, as reported in previous studies [7,8,9]. The environmental footprint of herbicides used on European farmland depends mainly on the activity of herbicides (herbicide companies), the environmental impact of active substances (environmental legislation and EU authorities) and a series of farmers’ decisions, who are the end users of herbicides. Previous studies have reported that it is difficult to obtain accurate data on herbicide use. However, pesticide sale statistics could be used under certain conditions, limiting their disadvantages [3,9,10]. Thus, national data of herbicide sale statistics available from the FAO since 1990 were used to assess the progress of herbicide use in the 21st century in the EU.

Another defining point of this effort that has attracted the attention of many authors is that the available information (herbicide sale/use of herbicides or other pesticides) does not equate to their exclusive use on agricultural land [6,9,10,11,12,13]. A small amount of pesticides are also applied to other areas such as lawns, gardens, railways, urban amenity areas, etc. In a recent US market survey of herbicide spending in different sectors, it was reported that 85% was spent on agriculture, 8% on industrial/commercial/government, and 7% on the home and garden sector [11]. Previous studies [12,13] estimated that approximately 10% of pesticides are applied to non-agricultural land, with herbicide application being dominant, in developed countries. Also, the US Pesticide Market Survey reported that among widely used pesticides, twelve (12) of the twenty-five (25) active substances (a.s.) were herbicides (glyphosate > Atrazine > Metolachlor-S > 2,4-D > Acetochlor > pendimethalin > Metolachlor > Propanil > Dicamba > trifluralin > Paraquat > Glufosinate), with glyphosate being the most commonly used [11]. In a previous study, Woodburn [14] showed that global glyphosate use, developed in the early 1970s, in non-crop areas was increased from 5.7% in 1994 to 8.1% in 1996. At the same time, an average annual volume consumption of 20% in recent years has been recorded in agricultural use. In 2008, glyphosate was reported as the main herbicide applied in non-agricultural land in seven European countries [15]. Eight years later, this herbicide continued to be the most widely used globally, with its application in agricultural and non-agricultural settings at 90% and 10%, respectively [16]. By 2020, the percentage of glyphosate sales used in the non-agricultural sector in the EU28 averaged 9%, while this varied greatly among EU countries, ranging from 2% in Austria to 40% in Belgium [6]. This variation in glyphosate use among the EU countries is mainly due to alternative weed control methods in national, regional, or local regulations and policies of herbicides use in non-agricultural/urban amenity areas [15]. Following the above studies, it is estimated that up to 9% of herbicides are applied to non-agricultural land in the EU. Since the European Union supports sustainable agriculture, understanding and evaluating the above agro-environmental parameters affecting herbicide use is expected to contribute to the reduction in pesticide use, through targeted agricultural policy measures.

Considering the aforementioned, this effort aims to assess the use of herbicides on EU agricultural land, with a comparative herbicide index (HI). Priority was given to the evaluation of the spatiotemporal distribution of herbicides, as well as to the distinction of the agro-environmental variables that influence this intensive use, among 27 EU Member States. The interaction of HI with key agro-environmental variables such as (a) agricultural land use, (b) farming systems, (c) farm size classification, (d) agricultural land cover types, (e) classification of herbicides by chemical family, and (f) long-term herbicide use was examined. Finally, a discriminant model was designed and tested that can incorporate the crucial indices shaping herbicide use in the 27 EU Member States. The interpretation of the model results can be used as a control/support tool in order to reduce herbicide use and their adverse environmental impact on EU agricultural land.

2. Data Analysis and Methods

2.1. Study Area

The European agricultural land of the 27 Member States (EU₂₇) was the study area. The following country codes for EU countries have been used in the text according to ISO 3166 [17] (except Greece, for which EL is recommended instead of the ISO GR code): Austria (AT); Belgium (BE); Bulgaria (BG); Croatia (HR); Cyprus (CY); Czechia (CZ); Denmark (DK); Estonia (EE); Finland (FI); France (FR); Germany (DE); Greece (EL); Hungary (HU); Ireland (IE); Italy (IT); Latvia (LV); Lithuania (LT); Luxemburg (LU); Malta (MT); the Netherlands (NL); Poland (PL); Portugal (PT); Romania (RO); Slovakia (SK); Slovenia (SI); Spain (ES); Sweden (SE).

2.2. Data Collection

Data of the agricultural sector for the 27 EU Member States were collected from three major statistical databases (FAOSTAT, Eurostat, and FiBL Statistics) for the period 2000–2021 and are described in Figure 1.

The selected agri-environmental indicators used in this study to assess herbicide use in the EU agricultural sector are analyzed below. About half (44%) of the world’s habitable land is agricultural land (AL) [18] used for permanent crops, arable crops, and livestock rearing in the form of land (pastureland—pmL) under permanent meadows and pastures and can be described as follows:

(i)

Based on land use, agricultural land (AL) can be divided into three (3) categories as follows (see list of abbreviations):

$$AL=arL+pcL+pmL$$

where pmL is land in long-term use (≥5 years) to grow herbaceous fodder, forage crops, through cultivation (sown) or naturally (self-seeded). However, the sum of pcL (land under permanent crops) and arL (arable land) makes up the crop land area (cropL), so

$$AL=cropL+pmL$$

In the calculations, the ratios of each agricultural land use (ha) to the total agricultural land (ha) were used, forming the I_cropL (cropland index) and the I_pmL (pastureland index), respectively.

(ii)

Based on farming systems, AL (agricultural land) can be divided into 2 categories as follows (see list of abbreviations):

$$AL=convAL+orgAL$$

In the calculations, the ratios of each farming systems (ha) to total agricultural land (ha) were used, forming the I_convAL (conventional AL index) and the I_orgAL (organic AL index).

(iii)

Based on land cover of cropland (cropL) where it is also the main recipient of herbicides due to their intensive use, it was divided into eight (8) categories (land cover groups or crop groups) as follows (see list of abbreviations):

$$cropL=G_{Cl}+G_{Dp}+G_{Rc}+G_{Ic}+G_{Pg}+G_{Fv}+G_{Pc}+G_{Other}$$

According to the structure of the EU₂₇ crop production, seven principal crop groups (Cl, Dp, Rc, Ic, Pg, Fv, Pc) were observed, as shown in Figure 1; another crop group (οther) was added as it emerged after subtracting data on the area of cultivated land from FAOSTAT minus Eurostat data [4].

In the calculation, the ratios of each crop group (ha) to the total cropland (ha) were used, forming the crop group index (IG) for each category (IG_cl, IG_Dp, IG_Rc, IG_Ic, IG_Pg, IG_Fv, IG_Pc, IG_Other).

(iv)

Based on farm size (FS) of cropland can be divided into 8 categories (Figure 1) as follows (see list of abbreviations):

$$cropL={FS}_1+{FS}_2+{FS}_3+{FS}_4+{FS}_5+{FS}_6+{FS}_7+{FS}_8$$

In the calculations, the ratios of each member of the farm size group to the total cropland (ha) were used, forming the farm size index (I_FS) for each category (I_FS1, I_FS2, I_FS3, I_FS4, I_FS5, I_FS6, I_FS7, I_FS8).

(v)

Based on herbicide chemical families (HCFs), the active substances (a.s.) with different modes of action and herbicidal activity where they are widely used in agricultural land (AL) can be divided into 10 categories (Figure 1). Total herbicide use (in kg of a.s.) is the sum of the quantities of each HCF class (kg of a.s.) and can be described as follows (see list of abbreviations):

$${HCF}_{total}=H_{phy}+H_{trz}+H_{amd}+H_{car}+H_{din}+H_{sulf}+H_{urc}+H_{bpd}+H_{urea}+H_{other}$$

In the calculations, the ratios of the quantity of each HCF class to the total quantity of herbicides were used, forming the herbicide chemical family index (I_HCF) for each category (I_Hphy, I_Htrz, I_Hamd, I_Hcar, I_Hdin, I_Hsulf, I_Hurc, I_Hbpd, I_Hurea, I_Hother).

(vi)

An attempt was made to create a comparative indicator of herbicide use in the EU conventional agricultural land (convAL), using the following equation:

$$Hebicides sales \left(\mathrm{k}\mathrm{g}\right) in convAL=Total hebicide sales \left(\mathrm{k}\mathrm{g}\right)\times 0.91$$

where 0.91 or 91% of total herbicide sales is the estimated herbicide amount applied in the conventional agricultural land (convAL) or at 39% of the total land area of the EU (based on 2016 land area Eurostat data); the rest (9%) was estimated as being used for urban or other purposes in the EU [6,12,13,15,16]. Finally, the herbicide index could be expressed by following equation:

$$HI \left(\mathrm{k}\mathrm{g}{\mathrm{h}\mathrm{a}}^{-1}\right)={\displaystyle \frac{Total hebicide sales \left(\mathrm{k}\mathrm{g}\right)\times 0.91}{ convAL \left(ha\right)}}$$

The conventional area of agricultural land (convAL) as the denominator of this equation was based on the fact that the use of chemical herbicides is prohibited on organically cultivated land [12].

2.3. Statistical Analysis

To evaluate the collected data, descriptive statistics were used as described in a previous analysis by Triantafyllidis et al. [9]. The statistical package SPSS version 26 was used as well as the MANOVA (multivariate analysis of variance) with Pillai’s Trace and Wilks’ Lambda indices of the multivariate testing.

Using solely the significant independent variables, linear discriminant analysis (LDA) was applied to explore the differentiation potential of EU countries by creating a discriminant model. The EU countries were treated as the grouping variables, while the determined parameters were taken as the independent variables. The mathematical model expressing the LDA in n-selected variables consisting of i dividing a linear function (F) is of the following form:

$$F_i=(c{i}_1{v}_1+c{i}_2{v}_2+\cdots +c{i}_n{v}_n)$$

where v₁ … v_n are the values of each variable that were examined and ci₁, ci₂, … ci_n are the correlation coefficients among variables [19].

3. Results

3.1. Spatiotemporal Distribution of the Herbicide Index (HI) in the Agricultural Sector of the European Union

The mean estimated value (n = 594) of the herbicide index (HI) across the European Union was 0.66 kg ha⁻¹ from 2000 to 2021 (Table S1). Despite the decline in conventional agricultural land, no (p > 0.05) differences were observed in the HI mean values among the years across the EU (Table S1); the trend of the HI index during the period 2000–2021 is described in Figure 2. Across the EU, the results showed that the agri-environmental indices IG_Rc and IG_Fv (root crop and fresh vegetable cultivation), I_FS7 and I_FS6 (farms with area from 30 to 99.9 ha), and I_Hcarb (carbamate herbicides) had a higher positive correlation (0.63, 0.53, 0.47, 0.41, 0.51, respectively) with the HI index (Table S2). The above results suggest that in root crop (Rc) and fresh vegetable (Fv) cultivation and on farms with an area of 30 to 99.9 ha, herbicides are widely used, while the widespread use of carbamide herbicides also increases the HI index. On the contrary, the agri-environmental indices IG_Cl (Cereal crops), I_FS2, and I_FS3 (farms with area from 2 to 9.9 ha) had higher negative correlation (−0.22, −0.24, −0.23, respectively).

After examining the geographical distribution of HI at the country level (Figure 3) (p < 0.05), differences were observed among EU Member States (Tables S3 and S4). Referring to the period from 2000 to 2021, the highest average values with S.D. of the HI in kg ha⁻¹ were observed in Belgium (2.23 ± 0.80), the Netherlands (1.52 ± 0.13), Cyprus (0.99 ± 0.29), Germany (0.94 ±0.09), France (0.92 ± 0.12), and Denmark (0.89 ± 0.26) (Tables S3 and S4). On the other hand, the lowest mean values with S.D. of the HI in kg ha⁻¹ were observed in Romania (0.24 ± 0.02), Greece (0.26 ± 0.09), and Bulgaria (0.28 ± 0.16) from 2000 to 2021 (Tables S3 and S4). The results showed that the distribution of the HI was independent of the geographical position of the European countries, such as from north to south or from west to east in the EU. It seems that other agro-environmental parameters also influence herbicide use in European Union countries. To assess the causes of this variability, selected agri-environmental parameters (formatted into indices to be comparable) were investigated, emphasizing the structural characteristics of the agricultural sector of each EU country.

3.2. Assessment of Selected Agri-Environmental Indices

After examining all selected agri-environmental indicators for the period from 2000 to 2021, the results showed that there are differences between EU Member States (Tables S4–S28), which suggests that the agricultural sector in the EU is not uniform. Regarding the land use of agricultural areas, the results showed that both the cropland index (I_cropL) and pastureland index (I_pmL) were not uniform among the 27 EU Member States (Figure 4). Differences were observed (Tables S5 and S6); the lowest values of I_cropL were observed in Ireland (0.10), Slovenia (0.39), Luxemburg (0.49), and Greece (0.50). As shown in the above countries that had a low cropland index (or had more pastureland), the use of herbicides (HI) was also low (Tables S4–S6). Across the EU₂₇, the estimated (n = 594, from 2000 to 2021) mean values for I_cropL and I_pmL were 0.68 and 0.32, respectively (Figure 4).

According to the cropping systems followed by farmers, who are the end users of agrochemicals, the results showed that among the 27 EU Member States, differences were observed (Tables S7 and S8). The highest organic agricultural land index (I_orgAL) values were observed in Austria (0.203), Sweden (0.134), and Estonia (0.126), while the lowest values were observed in Malta (0.002), Bulgaria (0.010), and Ireland (0.012). Although in some EU countries organic agricultural land occupied more than 10% [such as Austria (20.3%), Sweden (13.4%), and Estonia (12.6%)], in the EU, the mean organic agricultural land occupied only 6.0% for the period of 2000–2021 (Figure 5). Nevertheless, in 2021, the mean organic agricultural land occupied about 10% (I_orgAL = 0.100), showing a linear increase (R² = 0.9929) in the I_orgAL index over the period from 2000 to 2021 (Figure S1, Table S9). During the twenty-first century, both the increase in organic agricultural land and the decrease in conventional agricultural land of approx. 17.7% in 2021 compared to 2000 (Table S1) in the EU do not appear to affect HI values (Table S3).

Regarding land cover of cropland by main crop groups (Figure 6), the mean (n = 594) values of eight crop group indices (I_G) during the period 2000–2021 showed that the crop group index IG_cl, “cereals for the production of grain including seed (Cl)”, had the highest mean value in the EU₂₇ (Table S18). However, differences were observed among all eight crop groups in the 27 EU Member States (Tables S10–S18).

The estimated farm size index (IFS) in the European Union was not uniform, as shown in Figure 7. Farms with size up to 9.9 ha (FS1–3) had a mean index value of 0.175 (IFS1–3), while farms with size over 10 ha (FS4–8) had a mean index value of 0.825 (IFS4–8), as shown (Table S19). Meanwhile, differences in mean IFS values (n = 160) were observed among all eight farm size groups in the 27 EU Member States (Tables S20–S27). The highest mean values of the farm size index (IFS1–3) were observed in Malta (0.805), Slovenia (0.467), Cyprus (0.411), Romania (0.422), and Greece (0.391), while the lowest mean values of IFS1–3 were observed in Czechia (0.014), Luxemburg (0.016), France (0.020), and Denmark (0.024), as shown in Tables S20–S22.

Also, the estimated herbicide chemical family index (I_HCF) varied among the ten different chemical families (nine initial herbicide chemical families (HCFs) plus a HCF group of “other herbicides” as shown in Figure 8. Across the EU₂₇, the highest mean value (0.560) of this index (I_Hother) was observed from the HCF group of “other herbicides”, followed by phenoxy hormone products (phy) with a mean value of 0.140 for I_Hphy and amides (amd) with a value of 0.122 for I_Hamd, while the lowest mean value (0.0002) for I_Hurc was observed in the uracil chemical family, as shown in Table S28. However, statistical analysis of chemical pesticide families at the EU country level, due to a lack of values within the respective years, does not allow reliable conclusions to be drawn.

3.3. Assessment of the Herbicides Use in the 27 EU Member States

The qualitative criteria of MANOVA (multivariate analysis of variance), namely Pillai’s Trace = 7.273 (F = 27.698, df = 338, p = 0.000) and Wilks’ Lambda = 0.000 (F = 68.942, df = 338, p = 0.000), suggested that there was an impact of the 27 EU Member States on the examined variables (agri-environmental indices) (

Table 1. Results of linear discriminant analysis (LDA) to determine the correlation of the significant variables indicated by MANOVA among the 27 Member States of the EU.

Structure Matrix
Functions	1	2	3	4	5	6	7	8	9	10	11
Eigenvalue	40.616 a	21.954 a	17.170 a	10.577 a	4.458 a	2.124 a	1.419 a	1.186 a	0.605 a	0.345 a	0.105 a
% of Variance	40.4	21.8	17.1	10.5	4.4	2.1	1.4	1.2	0.6	0.3	0.1
Cumulative %	40.4	62.2	79.3	89.8	94.2	96.4	97.8	99.0	99.6	99.9	100.0
Canonical Correlation	0.988	0.978	0.972	0.956	0.904	0.825	0.766	0.737	0.614	0.507	0.308
Variables	Correlation between each variable and any discriminant function
IpmpL	−0.953 *	0.155	0.000	0.051	0.113	−0.003	−0.02	0.117	0.167	−0.022	0.103
IcropLb	0.952 *	−0.155	0.000	−0.05	−0.113	0.004	0.02	−0.118	−0.167	0.022	−0.103
IGFv	−0.028	0.697 *	0.318	−0.014	0.249	0.192	−0.026	0.549	0.067	−0.08	−0.026
IGPc	−0.093	−0.098	0.833 *	0.249	0.177	−0.127	0.071	−0.208	−0.177	−0.286	0.131
IGPg	0.016	0.019	0.000	−0.573 *	−0.275	−0.097	0.461	0.181	−0.164	−0.514	0.22
IGCl	−0.021	−0.143	−0.464	0.353	0.646 *	0.39	0.207	0.011	−0.071	−0.144	0.018
IorgALb	0.009	−0.067	−0.053	−0.124	0.434	−0.736 *	−0.096	0.244	−0.417	0.073	0.000
IconvAL	−0.009	0.067	0.053	0.124	−0.434	0.735 *	0.096	−0.244	0.417	−0.073	0.001
IGIc	0.018	0.11	−0.286	0.602	−0.25	−0.613 *	0.111	0.116	−0.261	−0.099	0.026
HI	0.061	0.326	−0.011	−0.072	−0.029	−0.157	0.609 *	−0.501	−0.246	0.404	−0.118
IGRc	−0.002	0.5	−0.057	−0.319	−0.126	0.134	−0.309	−0.565 *	0.295	−0.219	0.248
IGDp	−0.025	−0.053	−0.025	0.012	0.128	0.04	−0.288	0.000	−0.894 *	0.282	0.127
IGoth	0.045	−0.061	0.122	−0.091	−0.27	−0.002	−0.452	−0.008	0.424	0.711 *	−0.099

^a Eleven (11) canonical discriminant functions were used in the analysis. ^b This variable was not used in the analysis. * Pooled within-group correlations between discriminating variables and standardized canonical discriminant functions: largest absolute correlation between each variable and any discriminant function.

). Thirteen (13) of the thirty one (31) tested variables (agri-environmental indices) had values of p < 0.05 and were then subjected to linear discriminant analysis. The summary results of the LDA processing showed that all 594 cases (data values), which refer to the values of the variables that distinguish the values within the groups, were 100% valid.

Eleven DFs (1st–11th) were formed by the results of the LDA. All (11) DFs accounted for 100% of the total variance. However, the first four DFs could explain 89.8% of the total variance. Thus, the first DF accounted for 40.4% of the total variance and had the highest eigenvalue (40.616) and canonical correlation (0.988). The second DF had an eigenvalue of 21.954 and a canonical correlation of 0.978, while it accounted for 21.8% of the total variance. The third DF had an eigenvalue of 17.170 and a canonical correlation of 0.972, accounting for 17.1% of the total variance. The fourth DF had an eigenvalue of 10.577 and a canonical correlation of 0.956, accounting for 10.5% of the total variance.

The eigenvalues of the DFs are an indicator of how well the function differentiates the initial groups (27 Member States of the EU) [20]. In Figure 9, a clear separation of the 27 Member States of the EU is shown based on the considered variables. The classification rate was 93.9% using the original method and 92.4% using the cross-validation method. The obtained classification rates were 100% for 17 of the 27 EU Member States (FR, DE, IE, LU, NL, AT, HU, PL, RO, SI, DK, FI, SE, EL, IT, MT, PT); 95.5% for BE; 90.9% for ES, CZ, and CY; 86.4 for BG; 81.8% for LT; 77.3% for HR; 72.7% for LV; 59.1% for SK; and 50% for EE (Table S29a,b). Figure 9 also illustrates the group centroid values. It is noteworthy that the group centroid values are considered for the estimation of the classification ability of the LDA model and refer to the unstandardized canonical DFs evaluated based on group means. Fisher’s linear discriminant function coefficients for each of the 27 EU Member States are shown in Table S30.

4. Discussion

Short-term estimates are not suitable for monitoring and comparing fluctuating pesticide use [9,21]. For this reason, data from the agricultural sector of the EU 27 Member States (FAO, Eurostat, FiBL statistics) [4,22,23] during the first two decades of the 21st century were used to assess the HI of conventional agricultural land (convAL). The mean value of the HI was estimated to be 0.66 kg ha⁻¹ in the EU (Figure 3). However, this high value of the HI was lower compared to the estimated fungicide index (FI: 0.79 kg ha⁻¹) as determined in a recent study [9]. Despite the changes, concerning the climate or other factors, no differences have been observed in the HI from year to year in the European Union over the last 22 years (Table S1). The results show that farmers usually apply conventional weed control strategies, using chemical herbicides intensively and for a long time to avoid crop yield losses. However, optimizing herbicide performance and using good cultivation practices, such as crop rotation, the cultivation of competing crops/varieties, seed rate and sowing time control, formulation/adjuvants, and application techniques, could contribute to the reduction in herbicide doses [24,25,26,27,28]. Additionally, the use of non-chemical methods of weed control in organic farmland areas and beyond minimizes the risk of side effects to the agri-environment [29,30]. Although the organic agricultural land index (I_orgAL) showed a linear increase (Figure S1), the HI remained relatively stable in the period from 2000 to 2021. This is probably due to the fact that organic land covers only 6% of the EU’s agricultural land (Figure 5). Several studies have shown that agri-environment schemes and organic farming do not always deliver the expected benefits [31,32,33]. However, the organic agriculture area was increased in 2020 and 2021, occupying 9.6% and 10.0%, respectively. This finding indicates that this farming system could have a positive environmental impact on the HI in the future (Table S9).

Geiger et al. [34] supported the urgent need to restore biodiversity in Europe by minimizing the use of pesticides in cropland and especially in intensive crops where large areas are cultivated. The results of this study showed that agricultural land use has an impact on HI formation; it appears that in pastureland (pmL), the HI tends to be decreased, while in cropland (cropL), the HI is increased, probably due to intensive cultivation practices. It is obvious that cropL is the area where herbicides are mainly used. Nevertheless, the results showed that the claim that the higher the cropland index (ratio of cropland to agricultural land), the more herbicides are used is not a rule. Specifically, among the 27 EU Member States, pastureland covered more than 40% of agricultural land in 7 of the 27 countries, such as Ireland (89%), Slovenia (61%), Luxembourg (51%), Greece (50%), Austria (48%), Portugal (48%), and the Netherlands (43%) (Table S5). However, the HI in these countries was not lower in all cases compared to the other 20 EU countries (Figure 3). As it turns out, there are other factors that have a great influence on HI formation and need to be taken into account, such as land cover by crops.

Land cover by crops appears to have a role in herbicide use in the EU agricultural sector. Our results showed that mainly the crop groups “root crops (Rc)”, “fresh vegetables (including melons) as well as strawberries (Fv)”, “the plants harvested green from arable land (Pg)”, and “the industrial crops (Ic)” were the crop groups which had a negative impact on the HI over the last two decades in the EU₂₇ (Table S2). In contrast, the crop groups G_Cl, G_Dp, and G_Pc (see abbreviations) reduced the HI, showing their positive environmental impact. The results of this study suggest different effects on HI formation by the different crop groups, which is consistent with previous studies [16,35,36]. It appears that higher amounts of herbicide were mainly applied to high-value crops to control weeds and reduce economic losses. Similar results were reported by Berkhout and van Bruchem [37], stressing that the amounts of herbicides applied on potato and onion fields were three times and twice as much compared to arable land, respectively. Also, vegetable growers in Bangladesh applied more pesticides to achieve high yields and improve income [35]. Geiger et al. [34], studying the impacts on biodiversity due to agricultural intensification, showed that the number of wild plant species declined as the use of herbicides on cultivated land in cereal crops increased. As shown in this study, any changes in land cover affect weed management in turn. In particular, weed management using herbicides is carried out in high-value crops (e.g., root crops, fresh vegetables) with the aim of high yield per hectare and increased income. In the above land cover crop groups, weed management through sustainable agricultural systems or cultivation practices took place in a small part; on the contrary, weed control in the above crops seems to be treated either with frequent applications or with an increased dose of herbicides. The results reinforce the view that agricultural intensification to increase yields has disproportionately negative environmental impacts. However, although the use of herbicides per hectare was lower in the “arable land under cereals (including seeds)” crop group than in other crop groups, the larger area covered by this crop group resulted in a greater use of herbicides across the EU₂₇; similar results for pesticide use have been reported [38].

Also, among farm size classes (FS_1–8), the highest use of herbicides was observed in farms FS₆ and FS₇ (FS₆: 30–49.9 ha; FS₇: 50–99.9 ha), which covered about 25% of total cropland (cropL) in the EU (Tables S2 and S19). In contrast, the lowest use of herbicides occurred on farms FS₂ and FS₃ (FS₂: 2–4.9 ha; FS₃: 5–9.9 ha), which covered about 13%, from 2000 to 2021. The results denote the role of farm size classes in the sustainable use of herbicides. On the other hand, the differences in the application dose between the chemical families of herbicides [39,40] indicate their crucial role in the use of herbicides per hectare in the 27 Member States of the European Union; this effect of herbicide chemical families on the use of herbicides was observed. Carbamate herbicides have the strongest correlation (Table S2) with the HI compared with other chemical families. Piel et al. [41] reported that some a.s. of the carbamate herbicides such as propham and chlorpropham have been mostly used on potato crops as herbicides but also as plant growth regulators. Chlorpropham has been widely used until recently in root crops (potatoes, carrots, onions) and fresh vegetables such as lettuce to control annual and perennial broadleaved weeds and annual grasses [42]. Also, phenmedipham was used as a post-emergence herbicide on sugar beet/fodder beet, as proposed by the applicants. Moreover, cycloate is a carbamate herbicide which has been used only on beets, a crop particularly dependent on herbicide treatments (on average, in France, 15 of the 16 annual pesticide treatments on this crop are with herbicides) [41]. Nevertheless, the most used herbicide chemical family group was the “other herbicides chemical families” across the EU₂₇ from 2000 to 2021 (Table S28). The results showed that the use of this group of chemical herbicides, “other”, represents about 50.6% of the total use of herbicides in the EU; the highest use (84.4%) of this group occurs in Greece, while in 16 of the 25 countries (in 2 countries, no data were available), use of this group of herbicides accounted for more than 50% of total herbicide use on conventional EU farmland (Figure 8). Many chemical families belong to this group (HCF_other), such as organophosphorus, nitrile, anilide, cyclohexanedione, chloroacetanilide, pyridyloxyacetic-acid, aryloxyphenoxyproprionic, thiadiazine, benzoic-acid, pyridinecarboxylic-acid, thiocarbamate, imidazolinone, triketone, and diphenyl ether herbicides. However, the ratio of glyphosate—an organophosphate—compared to all herbicides in 2017 (%) was 34% in the EU₂₈ as reported by Antier et al. [6]; this explains the high values observed in the herbicide chemical families index in the group “other herbicidal chemical families”. However, the extensive use of herbicides imposed selection pressure on weed communities, and as a result, the first herbicide-resistant (HR) weed populations emerged [43]. Farmers often apply herbicides at doses higher than those recommended in order to control HR weeds [44], thus perpetuating this vicious cycle. Currently, more than 260 weed species have been reported globally to evolve resistance to more than 160 herbicides [45], including glyphosate, the most successful synthetic herbicide [46]. In addition to the emergence of noxious HR weeds and their limited control options by farmers, the extensive use of herbicides has adverse effects on the agri-environment.

After examining the above agri-environmental indices, an attempt was made to combine their structural variability in each of the 27 EU Member States. Using only the significant independent variables, linear discriminant analysis (LDA) was applied to explore the differentiation potential of the 27 Member States of the EU by creating a discrimination model. Each EU country was treated as a clustering variable, providing the grouping variables, while the determined parameters were taken as the independent variables. Linear discriminant analysis (LDA) of 11 variables was examined to describe this differentiation of the 27 EU Member States (Table 1). The examined agri-environmental indices were related to agricultural land use (I_pmL), farming system (I_convAL), land cover of cropland (IG_Fv IG_Pc IG_Pg IG_Cl IG_Ic IG_Rc IG_Dp IG_oth), and the HI, and pooled within-group correlations between the discriminant variables and the standardized normal discriminant functions (DFs) were able to explain this variability across the 27 EU Member States. The highest absolute correlation between each variable and any discriminant function is shown in Table 1.

The results in Table 1 and Table S2 showed that agro-environmental indices (such as agricultural land use, farming systems, land cover by different crop groups, farm size, and different herbicide chemical families) play a decisive role in the intensity of herbicide use, forming the HI across the European Union over the period 2000–2021 (Table S2). Meanwhile, these results, combined with their structural variability in each of the 27 Member States as described by Fisher’s linear discriminant functions (Table S30), could be an effective management tool to reduce the herbicide use. Finally, the assessment of the above eleven agri-environmental indices clearly discriminates the herbicide use in the 27 EU Member States, creating an effective tool to estimate use and further improve herbicide management in the EU₂₇.

5. Conclusions

The herbicide index (HI) calculated in this study showed that herbicide use was 0.66 kg a.s. per hectare of conventional EU agricultural land in the period from 2000 to 2021. However, the HI varied between the 27 EU Member States. This variation appears to be due to the fact that the EU agricultural sector is not uniform. According to the land use of agriculture area, cropland in contrast with pastureland has a negative impact on the formation of the HI, suggesting that the larger the cropland area, the higher the use of herbicides in the EU₂₇. Nevertheless, the results showed that the claim that the higher the cropland index (ratio of cropland to agricultural land), the more herbicides are used is not a rule. As it turns out, there are other factors that have a great influence on HI formation and need to be taken into account, such as land cover by different groups including crops. The land cover by “root crops” and “Fresh vegetables (including melons) and strawberries” (high-value crops) suggested a high negative impact on the herbicide index. Also, farm size is a crucial variable in HI formation across the EU₂₇; the greatest use of herbicides was observed in farms FS₆ and FS₇ (FS₆: 30–49.9 ha; FS₇: 50–99.9 ha), which covered about 25% of the total cropland (cropL) in the EU. In parallel, the herbicide chemical family of “carbamates” had a higher negative impact on the formation of the herbicide index (HI) compared to the other chemical families. Using only the significant independent variables (13 out of 31), linear discriminant analysis (LDA) was applied to explore the differentiation potential of the 27 Member States of the EU by creating a discrimination model. The knowledge of the main parameters of the agricultural sector as well as their contribution to the use of herbicides in the European Union and their geographical distribution is expected to be an effective management tool to reduce the use of herbicides, to improve the management of weeds in agricultural land, and to reduce their environmental impact in the EU.