The study was conducted in the area of Montes Orientales, 30 km north of the city of Granada (Spain), for being one of the areas most intensely dedicated to traditional olive-orchard cultivation in the province of Granada (Guzmán & Alonso, 2004a). The study area occupies some 20 km from east to west and some 18 km north to south. This is a rather homogeneous zone, with altitudes of 960 to 1130 m asl. The climate is extreme continental Mediterranean, with long, cold winters, and equally long, warm summers. The mean precipitation does not reach 600 mm annually, rain falling mostly in the months of October to May. The main crop is olive, predominantly the cultivar ‘Picual’, grown under conventional as well as intensive systems. Also, cereals and other herbaceous crops abound in the area.

We considered three types of management when selecting the sampling stations, seeking uniformity regarding geographical and geophysical characteristics of the different olive-orchards.

Olive orchards under organic production (ORG hereafter), following EC Regulation no. 2092/91 of 24 June 1991, used environmentally friendly agricultural practices based on environmental and economic sustainability as well as self-regulation of the trophic chains present in the agroecosystem. In this sense, no agrochemicals were used nor was the soil ploughed. During the study, only one treatment of Bacillus thuringiensis Berl. was applied between May and June, depending on the orchard.

In the olive orchards under conventional management, pesticide and herbicide use was widespread and routine, though two types of management were distinguished according to ploughing depth: strict conventional (CON hereafter), and conventional non-tillage (NT hereafter). In the latter, at most, shallow tillage was applied under the tree canopy at the beginning of spring. The agricultural characteristics of the different management types are listed in Table 1.

Sampling was conducted for three consecutive years (2007, 2008, and 2009), between May and August, depending on the year, as the time when the largest differences are observed between management types (Redolfi et al., 1999) and which has the highest arthropod abundance (Ruano et al., 2004; Santos et al., 2007a; Jerez-Valle et al., 2014). The number of samples was reduced progressively from one year to the next, in the light of the results of accumulation curves for each campaign. In 2007, we selected 15 sampling stations, 5 for each type of management and sampled fortnightly on five dates between early June and early August. In 2008, we selected 12 sampling stations, 4 for each management type, taking samples every two weeks from May to July on four different dates. Finally, in 2009, we also selected 12 stations, sampling fortnightly but only on three different dates, between late May and June.

Each sampling station was composed of a row of four trees sampled, leaving one alternately unsampled. In this way, the distance between sampled trees was greater than 15 m, to ensure the independence of the data. In addition, the minimum distance between each station was 500 m.

We sampled the canopy by the beat-down technique (Wilson, 1962) for 40 seconds, beating four branches per tree (one per orientation) taken at random. We used a 45-cm-diameter sweep net with a rod 55 cm in length. The content of the net (methodological sample) was emptied into plastic bags properly labelled, with a few drops of ethyl acetate as an anaesthetic to avoid predation. In the laboratory, all samples were stored in a freezer at -20ºC until processing. The sampling unit was considered to be the average of four methodological samples taken at each station per sampling date, to avoid dispersion of data and increase uniformity.

Each sample was processed in the laboratory, separating and classifying the arthropods. Under a binocular microscope, all the adult and juvenile specimens were classified to the level of order, except in the case of Euphyllura olivina (Costa) (Hemiptera: Psyllidae) and Prays oleae (Bern.) (Lepidoptera: Plutellidae), which, for being rampant pests were considered apart. The family Formicidae was also separated from the rest of Hymenoptera due its high number. In addition, within the order Hemiptera, the suborder Heteroptera was separated from the suborders Auchenorrhyncha and Sternorrhyncha, which were lumped under the name “Other Hemiptera”.

Summary statistics. Absolute abundance of arthropods (N) was considered as the number of individuals recorded in total per management type. In addition, for each management type, the mean number of individuals per sampling station () and standard deviation (SD) were calculated, differentiating samples taken on different dates by leaving at least 15 days between samplings in all cases. The median (Me) for each order and management type per year was also calculated as measure of central tendency, as well as the first and third quartiles (Q1 and Q3, respectively).

Due to the non-normality of the data, despite several transformations, the Kruskal-Wallis non-parametric test (K-W) was used in addition to the Mann-Whitney U test, for multiple and two independent samples, respectively, to detect differences at 95% significance level. We also performed a pairwise comparisons analysis (Nikolić, 2007) to determine between which values such differences occurred. The calculations were made using R software (R Development Core Team, 2012).

Accumulation curves. For each type of management, species-accumulation curves were constructed following the methodology of Jiménez & Hortal (2003). In this case, the curves refer to orders. Hemiptera was considered as two groups (Heteroptera and “Other Hemiptera”). The family Formicidae was included in Hymenoptera, while E. olivina in “Other Hemiptera” and P. oleae in Lepidoptera. The Clench fit was applied to each curve, by nonlinear estimation by the Simplex & Quasi-Newton method (StatSoft, 2005). To construct the accumulation curves from the real data, the program EstimateS (Colwell, 2009) was used, and the model was fit using the program Statistica 7.1 (StatSoft, 2005).

Meteorological data. Data from Agroclimatic Information Network of Andalusia (http://www.juntadeandalucia.es/agriculturaypesca/ifapa/web) and Crop Disease Warning Information Network (http://www.juntadeandalucia.es/agriculturaypesca/raif/) of the Junta de Andalucía were used to determine weather conditions on sampling days. We considered the stations located within 50 km from the centre of the study area (16 stations in total), recording the daily values of precipitation (mm), average temperature (°C), relative humidity (%) and daily radiation accumulated (MJ/m²). We performed a hierarchical cluster analysis using between-groups linkage method using the squared Euclidean distance, with data from each sampling date and for the previous four days. This provided a dendrogram classification with the results of the cluster. The calculations were performed using the R software (R, 2012).

Discriminant analyses. We built two discriminant functions to classify the olive-orchard stations, paying attention to the management type and considering the criterion “organic/non-organic”. Moreover, we used three different discriminant methods to determine the most efficient: Linear Discriminant Analysis (LDA) (Venables & Ripley, 2002; Ripley, 2008), Multiple Discriminant Analysis (MDA) (Hastie et al., 1994, 1995; Hastie & Tibshirani, 1996) and Support Vector Machine (SVM) (Fan et al., 2005; Chang & Lin, 2011).

In addition, two different analyses were developed based on the criteria for choosing the groups to include: analysis 1 (with all groups) and analysis 2 (with taxa that showed significant differences in the three years). Finally, looking for an improvement in classification rates, we considered meteorological criteria to divide the data set into two subsets of dates, which we analysed. All calculations were performed using the software R (R, 2012) with “package MASS” (Venables & Ripley, 2002), “package mda” (Hastie et al., 2011) and “package e1071” (Dimitriadou et al., 2011).

A total of 42,369 specimens were collected, belonging to 14 different orders (Heteroptera considered separately from other Hemiptera). In 2007, 25,602 specimens were collected, 11,275 in 2008 and in 2009, the highest abundance being of ORG (32,160 individuals within three years), followed by CON and NT (6,098 and 4,111, respectively).

By years, in 2007 the highest abundance was recorded in ORG with 22,990 individuals, followed by CON and NT (1,346 and 1,266 individuals, respectively). Significant differences were detected between ORG, NT, and CON management (Fig. 1a) with respect to medians (Me_ORG=207.8; Me_NT=12.0; Me_CON=12.0; K-W=49.5, p<0.05; mean abundance per station of each management type: _ORG=249.9 ± 18.3; _NT=13.1 ± 0.8; _CON=13.6 ± 0.9). The analysis was repeated removing pests of great abundance (Fig. 1b), confirming the differences found in the previous analysis (Me_ORG=14.0; Me_NT=8.8; Me_CON=5.8; K-W=26.1, p<0.05; mean abundance per station: _ORG=22.8 ± 1.7; _NT=11.3 ± 0.7; _CON=7.2 ± 0.5).

In 2008, abundance also proved highest in ORG with 6,879 specimens, followed by CON and NT (2,921 and 1475 individuals, respectively). Significant differences were detected, in this case between the three management types (Fig. 1c) with respect to the medians per station (Me_ORG=122.3; Me_NT=19.1; Me_CON=44.0; K-W=40.8, p<0.05; mean abundance per station: _ORG=125.2 ± 15.2; _NT=23.0 ± 2.6; _CON=45.6 ± 3.9). Repeating the analysis without pests (Fig. 1d), we detected differences between ORG and the other two management types (Me_ORG=14.8; Me_NT=10.0; Me_CON=10.9; K-W=9.3, p<0.05; mean abundance per station: _ORG=22.8 ± 3.3; _NT=11.3 ± 0.8; _CON=9.2 ± 0.4).

Finally, 2,291 individuals were collected in 2009 in ORG, this again being the management type with the highest abundance, followed by CON and NT (1,831 and 1,370, respectively). On this occasion, the only significant differences were between ORG and NT when considering the medians (Fig. 1e) (Me_ORG=47.3; Me_NT=24.9; Me_CON=32.6; K-W=10.2, p<0.05; mean abundance per station: _ORG=47.7 ± 2.3SD; _NT=28.5 ± 2.5; _CON=38.2 ± 3.4). When the pests were removed (Fig. 1f), no significant differences were found (Me_ORG=12.6; Me_NT=10.5; Me_CON=11.5; K-W=4.2, p > 0.05; mean abundance per station: _ORG=13.3 ± 0.8; _NT=11.2 ± 1.4; _CON=15.0 ± 1.3).

As expected, the arthropod fauna associated with the olive canopy was more abundant in ORG, even without considering the pests, except for 2009, where no significant differences were found when pests were not considered. These results support the idea that the arthropod community is strongly influenced by agricultural practices (Santos et al., 2007b). Agrochemical use and removal of vegetation cover are directly related to a decline in the abundance and diversity of beneficial arthropods (Ruano et al., 2001; Cárdenas et al., 2006; Porcel et al., 2013). Desneux et al. (2007) performed an extensive review concerning the sublethal effects of pesticides, emphasizing that chemicals affect not only the mortality of beneficial arthropods but also seriously reduce their neurophysiological, physical, and behavioural capacities.

Moreover, the median and the first and third quartiles of each group per methodological sample (sampled tree) were calculated according to the management type (Table 2), as well as the mean number of individuals and its SD (Table S1 [online supplement]). Most groups proved to be most abundant in ORG compared to the two other management types. Specifically, P. oleae, Heteroptera, and “Other Hemiptera” showed significant differences in this sense for the three years. E. olivina, Formicidae, Neuroptera, and Rhaphidioptera significantly differed in two of the three years, although ORG was not always the management with the greatest abundance. For example, Neuroptera was significantly lower in ORG for 2007 and 2009.

The pest E. olivina was dominant in all years, regardless of the management type, presumably due to their high biotic potential (Civantos, 1999). This psyllid seriously damages olive trees, especially increasing the percentage fall of inflorescences and reducing fruit set (Chermiti, 1992). Meanwhile, P. oleae sharply declined in 2008. Other studies on this lepidopteran have found considerable interannual fluctuations (Delgado & Cuesta, 1995; Hegazi et al., 2011) related to environmental variables. The high importance of taxa such as Coleoptera and Araneae in the olive grove is also remarkable (Ruíz & Montiel, 2000; Ruano et al., 2004; Santos et al., 2007b; Scalercio et al.; 2009; Rei et al., 2010; Cotes et al., 2011), hardly register significant differences in abundance among management types. However, members of these taxa have been proposed as bioindicators of olive management at lower taxonomic levels. Cárdenas et al. (2006) used spider families, while Cotes et al. (2009) used the Coleoptera morphospecies level, both cases proving useful for differentiating between organic and non-organic management.

Moreover, taxa with significant differences for at least two of the three years were invariably flying taxa typical of the canopy (Ruano et al., 2004; Santos et al., 2007b; Rei et al., 2010), or with a high capacity of movement (e.g. Formicidae). The case of Neuroptera is particularly striking, as it was more abundant in non-organic than in ORG management (in 2008, without significant differences). Other authors have reported similar results, concluding that the greater abundance of this group in conventional olive orchard is related to a very strong dominance of Chrysoperla carnea (Steph.), to the detriment of other species, which on the contrary are themselves more abundant in organic management (Ruano et al., 2001; Corrales & Campos, 2004; Porcel et al., 2013). C. carnea has a high capacity to develop resistance in the treated agro-systems (Zaki et al., 1999). Rhaphidioptera, despite its scarcity, was more abundant in ORG than in the other management types (in 2007, without significant differences). This is a rare group, whose individuals are entomophagous in all life stages; the larvae need a long development period, usually overwintering during the pupal stage (Aspöck & Aspöck, 2009). These biological characteristics make these insects highly sensitive to any treatment or alteration of the environment, and therefore ORG management provides them a more suitable habitat for survival.

Finally, the number of taxa was found to be higher in ORG than in the other management types, nor was any group found to be exclusive to any particular type of management. This result indicates that organic management is more stable than the rest, since over three years the same taxa were recorded, whereas in other management types, they fluctuated more sharply.

Furthermore, the median and the first and third quartiles of each taxon per methodological unit based on an organic/non-organic approach (Table 3) were calculated, as well as the mean and its SD (Table S2), as confirmed by the results. In this case, only Heteroptera, Formicidae, and Rhaphidioptera showed significant differences for three years with respect to medians, being more abundant in organic than non-organic management.

By constructing accumulation curves, we evaluated the sampling effort for each year (Fig. 2a-c). In all cases, the resulting slope was m<0.1 (Table 4). Furthermore, the total number of orders (TNO) which predicts the Clench adjustment proved to be very close to the observed number of orders (Sobs) so that the percentage of individuals recorded (FR%) was very high. In addition, the number of samples required for a slope of 0.1 (n(m=0.1)) was in any case far below the actual number of sample units used (n). The R² value (parameter directly proportional to the goodness of fit) was R²>96% for the three years.

Species-accumulation curves constitute a useful tool in biodiversity studies (Moreno & Halffter, 2000; Willott, 2001) since they i) give reliability to biological inventories and make them comparable, ii) estimate the sampling effort required for a reliable inventory, and iii) extrapolate the number of species in an inventory to estimate the total number of species in an area (Soberón & Llorente, 1993; Colwell & Coddington, 1994; Gotelli & Colwell, 2001; Jiménez & Hortal, 2003). In our case, the primary purpose of constructing accumulation curves was to evaluate the sampling intensity. All other considerations must be taken with caution, as the methodology is designed to use the taxonomic level of species rather than order (Jiménez & Hortal, 2003).

The results for 2007 helped reduce the number of sample units of 75 to 48 for the 2008 campaign, as the number of samples needed for a slope of 0.1 was found to be in any case much lower than that used. Something similar happened in 2008, thus reducing the number of sample units to 36 for 2009. Despite the steady decline in the number of samples, sampling intensity applied in the study was adequate, since the slope of the curve fit was in all cases above 0.1 and the percentage of fauna recorded reached very high values. Nevertheless, the number of significant differences between taxa detected in 2009 was the lowest of the three years, so that we do not recommend reducing the number of samples. In light of these results, the inventory of arthropod fauna at the level of order can considered quite complete.

Discriminant analysis was performed with LDA, using the 2007 data to classify the 2008 and 2009 separately, and 2008 to classify 2009. Subsequently, the weighted average percentage of success of these three tests was calculated. Two different data sets were used. In the first (analysis 1) we used all taxa while in the second (analysis 2) we considered only those groups with significant differences in the three years of sampling with respect to medians (Tables 2 and 3), at least under one of the two criteria (“management type” or “organic/non-organic”), regardless of the pest: Heteroptera, “Other Hemiptera”, Formicidae, and Rhaphidioptera.

Firstly, we assessed the ability of the discriminant analysis to differentiate between “management types” (Fig. 3a). The weighted average of success was 54.6% and 57.1% in analyses 1 and 2, respectively. However, when we considered the criterion “organic/non-organic” (Fig. 3b), the weighted average percentage of success increased to 86.6% (analysis 1) and 84.9% (analysis 2).

In light of these results, it may be desirable to use some other method to improve the classification rates found with the LDA method. In recent studies (Xiong & Cherkassky, 2005; Mazanec et al., 2008; Moreno & Melo, 2011), one of the classification methods proposed is the SVM, the key feature of which is that it does not require prior distributional assumptions of the data because it is a nonparametric method. A third method of classification is the MDA, which can be considered an extension of the LDA, in which mixtures of normal distributions are used to make a density estimation of each class.

However, in the case of MDA, the success rates were reduced in both “management type” (Fig. 3c) and “organic/non-organic” (Fig. 3d). Only one analysis of this test was able to improve the weighted-average percentage of accuracy to 87.7%. SVM achieved higher success rates in the two analyses, both “management type” (Fig. 3e) and “organic/non-organic” (Fig. 3f), but analysis 1 failed to improve the results. In general, none of the analyses proved satisfactory, due to the high number of failures recorded, especially in classifications of 2009.

Due to interannual differences registered and disparity of classification between years, we used agroclimatic data records RIA and RAIF to characterize the different sampling dates in search of meteorological analogies that would allow us to select the most similar ones. Two distinct groups resulted (Fig. 4): one joined the first sampling dates of each year (“Group A”) characterized by having a high average humidity, sparse and scattered precipitation, and moderate average temperature and radiation (Fig. S1 [online supplement]). On the other hand, another group was formed by last dates of each year (“Group B”), i.e. days with low relative humidity, scarce or null precipitation, high radiation, and higher temperatures. Given the strong differences detected, we repeated the discriminant analysis with these two groups of dates separately.

“Group A” (Fig. 5) recorded very low average success rates for all analysis-method combinations, in “management types” (Fig. 5a,c,e) as well as in “organic/non-organic” (Fig. 5b,d,f). The number of successes in 2009 continued to be lower than in 2008.

By contrast, “Group B” (Fig. 6) improved the classification rates, especially with analysis 2. In “management types” (Fig. 6a,c,e), all three methods improved the success rate with analysis 2. It is notable that the percentages of success in 2009 were very similar to those of 2008. Using “organic/non-organic” (Fig. 6b,d,f), we found very accurate results with analysis 2 for LDA (97.9% accuracy) and MDA (93.8% accuracy), both cases reaching 100% success in 2009 with respect to 2007 and to 2008. This result provides analysis consistency and reliability.

Discriminant analysis according to “organic/non-organic” proved to be far more effective than “management type”. In fact, in the latter, no satisfactory results were found for any of the analyses presented. Similar results have been reported by other researchers (Ruano et al., 2004; Santos et al., 2007b; Cotes et al., 2009), perhaps due to ecological practices involving a leap in quality compared to the conservation of the ecosystem in all its aspects (Parra et al., 2007). In fact, non-organic management types are more heterogeneous with respect to each other, depending on the judgement of each farmer.

The excellent results of “Group B” could be due to the meteorological homogeneity of the summer period, when weather conditions are more stable between different years. The spring term in the Mediterranean area in general is characterized by high atmospheric instability, causing numerous storm events, so that the community of arthropods from one year to another can vary greatly over this season. Cotes et al. (2011) found similar results, associating the improvement in classification with the pre-blooming and post-blooming period of the olive tree.

For this reason, for discrimination between management types and especially between organic and non-organic practices, we recommend the use of taxa that differ significantly in their abundance, excluding pests. In addition, we propose beginning sampling in the second half of June and continuing it to the end of July, attempting to sample in weeks with relatively stable and homogeneous weather conditions. Moreover, the inclusion of different discriminant methods did not quantitatively or qualitatively improve the classification, so that we propose that more attention should be placed on the sampling characteristics and the data set than on the statistical method.

The use of olive-canopy arthropods at the level of order proved to be useful in discriminating between management types, especially among organic and non-organic practices. The weather should be taken into account, as the selection of the most homogeneous dates is crucial in order to quantitatively and qualitatively improve success rates. However, the inclusion of different discriminant methods did not provide significant improvement. The consideration of taxa with significant differences for the three years gave stability and reliability to the study, despite the differences between years. Therefore, for discrimination between management types, we recommend sampling between the second half of June and late July to prevent unstable conditions in late spring and outliers in August. In addition, we propose using the discriminant analysis only for groups with significant differences, regardless of the statistical method.