Weed infestation and factors affecting weed incidence in spring cereals in Finland - a multivariate approach

Weed vegetation of spring cereal fields in southern and central Finland was analyzed by ordination methods to provide a community level description of weed populations. Attention was paid particularly to the relative importance of environmental factors affecting weed incidence such as crop management, soil properties and weather conditions. A data set of 33 weed taxa from 252 fields was subjected to both indirect and direct gradient analysis. Indirect ordination was obtained with correspondence analysis (CA), and direct gradient analyses were performed with redundancy analysis (RDA) and withcanonical correspondence analysis (CCA) relating environmental factors to the occurrence of weeds. Among several management factors, continuous herbicide use explained best the variation in the species composition of weed flora. Weed vegetation was also associated with soil type, moisture conditions and soil pHh 2oOrdination diagrams visualized the species-environment interactions and detected characteristic weed species for different geographical regions. In addition to ordination analyses of weed flora, the level and structure of weed infestation are described. The density of weeds averaged 170 plants m 2(median=l24) and the air-dry weight of weeds 320 kg ha I (median=lB3). The average weed density was the same in different soil types, but the weed biomass was lower in clay soils than in coarse mineral and organic soils.


Introduction
Arable fields are continuously subjected to differ- ent agricultural measures particularly in annual crops.Although many weed species are adapted to the prevailing conditions, the constantly changing habitat selectively affects weed communities and, consequently, changes the weed flora (Rade-MACHER et al. 1970, Reuss 1981, Mahn 1984, Chancellor 1985, Légére et al. 1993).
Weed flora in spring cereals was investigated during 1982-84 in Finland (Ervio and SALONEN   1987).Attention was paid particularly to the changes in weed infestation by comparing the data with the previous study from the 1960 s (Mukula   et al. 1969).The occurrence of individual weed species was related to several explanatory variables by the analysis of variance and regression tech- niques.These methods are appropriate if detailed responses of particular weed species to explanatory factors are studied.The problem was, however, to give a summary of the relative importance of fac- tors affecting the weed incidence.Therefore, the data from weed survey was subjected to ordination analyses which have proved to be appropriate for community level description of weed vegetation (Ter Braak 1987 a).
Multivariate analysis of community data is fre- quently applied in ecological studies to summarize the information in samples-by-species data matri- ces (Gauch 1982).In weed science, the multivari- ate approach is feasible to describe and predict the response of weed vegetation to farming practices (POST 1988).Multivariate methods in ecology can be divided into three groups (JONGMAN et al. 1987): direct gradient analysis (regression), indirect gradient analysis (ordination) and classification (cluster analysis).Indirect methods analyze the species data only, whereas species-environment interactions can be analyzed simultaneously by direct methods.
In this paper, the weed survey data from 1982-1984 was subjected to ordination analyses to give a community level description of weed flora in spring cereal fields.The objective was to find charac- teristic weed species in different geographical re- gions and to illustrateresponses of weed vegetation to environmental factors.Furthermore, the level of weed infestation, proportion of the most abundant weed species and the occurrence of weeds in differ- ent soil types are reported.

Material and methods
A total of 267 spring cereal fields (barley, oats or wheat) in southern and central Finland were studied  during 1982-1984.In each field there were 4 to 5 sample plots of 0.25 m in size from which the above-ground occurrence of 33 weed species (Table 1) or, in fact, weed taxa was assessed in late July by counting the number and weighing the air- dry biomass of weeds.The sample plots were not sprayed with herbicides.Frequency of weeds (Table 1) denotes the proportion of the fields where the particular weed species was observed out of the all fields studied.Detailed information of the survey and the occurrence of weed species has been given by Erviö and Salonen (1987).
Data on factors involved in each field was col- lected either by observing, measuring or by inter- viewing the farmer.Twelve factors describing either the current crop, crop rotation, soil properties or climate (Table 2) were used as environmental variables in the CCA.The factors were chosen from among the 21 factors studied in the regression ana- lysis and considered the most important (Erviö and  Salonen 1987).The survey localities were grouped into three regions based on their geographical locations: South-western Finland (SW), eastern part of central Finland (CE) and western part of central Finland (CW).
Features of regression analysis and ordination are integrated in canonical ordination techniques (Jongman et al. 1987).These techniques provide a direct analysis of species-environment interactions which was earlier possible only by regression methods.'Canonical correspondence analysis' (CCA) by Ter Braak (1986) is probably the most common canonical ordination technique currently applied in various ecological studies (Birks and Austin 1992).CCA and the related indirect technique 'correspondence analysis' (CA) (Gauch 1982) have been applied also in agricultural research (Jukola-Sulonen 1983,Wentworth et al, 1984,Post 1986,  Siepel et al. 1989,PvSek and LepS 1991, Dale et al. 1992).CA and CCA fit the unimodal curve to the species-environment data, whereas a linear response model between species data and environmental variables can be fitted by the 'redundancy analysis' (RDA).The ordination techniques mentioned above are all available in the computer program CANOCO (Ter Braak 1987 b).
Environmental variables were either qualitative (nominal scale) or quantitative (interval scale) (Table 2).The crop rotation was considered cereal dominant if a cereal crop had been grown at least for three years of theprevious four years.Otherwise it was classified as mixed rotation.The use of herbicides indicates only the intensity of chemical weed control, not the type of herbicides applied during the last nine years.The soil pHh 2 0 was measured from the top 0-20 cm layer.The soil type of fields was classified into three categories: clay (clay content >30%), organic (>20% organic mat- ter) and coarse mineral soils.The subjective assess- ment of soil moisture was primarily based on the soil type and the drainage of the field.Nominal type environmental factors were transformed into binary dummy variables.Due to missing values of ex- planatory factors, some sample fields had to be excluded, since missing data are not accepted in the CANOCO run.Thus, a final data set consisted of Table 1.Frequency, the effective number of occurrences (N 2) and average biomass production of the 33 weed species studied in 252 spring cereal fields.Frequency denotes the proportion of the fields where the species was found.The N 2 value obtained from the CANOCO run is based on the weighted averages of weed densities and it indicates the number of fields where the species was abundant.Air-dry biomass indicates the average infestation of the species in those fields it was found.

Weed taxa
Code" Frequency N 2 Biomass % g m 2  252 fields.The geographical regions were used as ordination axes.Only the central area of the dia- environmental variables in RDA, and as covari- gram is shown to improve the visibility of species ables in partial CCA.near the origin.Consequently, some species and Ordination analyses were performed with the environmental variables lie outside the drawn area CANOCO program (Ter Braak 1987 b) applying (Figs. 4 and 5).
CA, CCA and RDA.Ordination diagrams (species-Due to the skewed distribution of the response environment biplots) were drawn with the CANO- values (weed density and weed biomass) the weed DRAW program (Smilauer 1990).The relation- data was log-transformed (ln(y+l)) in the ship between the weed communities and environ- CANOCO run.Species diversity was described by mental variables is displayed with the first two the N 2 value from the CANOCO output.The N where X is Simpson's diversity index, nj is the number of ith species in the population and N is the total number of all S species in the population.

Results
Occurrence of weeds 2 The weed density averaged 170plants m (SE= 10, median=l24) and the biomass production 320 kg ha' 1 (SE=23,   183).The total weed biomass correlated weakly (r=0.54,P<0.01) with the total weed density.A typical weed density was 50-150 weeds m' 2 , whereas the biomass production was distributed more evenly into different classes (Fig. 1).Weed densities in different soil types were at the same level, but the biomass production of weeds was on average lower in clay soils than in coarse mineral or organic soils (Fig. 2).
Only ten weed species occurred in more than half of the fields studied (Table 1).The ranking order based on the N 2 value was slightly different from the frequency order.The N 2 value for samples averaged 6.9 (range 2.1-13.4),i.e. on average there were seven relatively abundant weed species in each field.
Moreover, in order to emphasize the relative importance of different weed species, they were ranked according to their average biomass production (Table 1), and also with regard to their proportion of the total density and biomass of weeds (Fig. 3).The nine most dominant weeds constituted two- thirds of the total weed infestation.
value is analogous to Hill's N 2 diversity number (Hill 1973).For samples, N 2 is the inverse of Species-environment relationships CCA and RDA were applied to the species ordina- tion in the three geographical regions.Both tech- niques characterized the typical weed species of different regions illustrated here by the RDA dia- gram (Fig. 4) which provided a slightly better separation of samples and weed species than the CCA diagram.
The CCA ordination diagrams for the density and biomass data were very much alike.Thus, only the diagram for biomass data is shown, since it provided somewhat higher eigenvalues (Table 3, Fig. 5).Eigenvalue denotes the dispersion of the species scores along the ordination axis, and is thus a measure of importance of the ordination axis (JONGMAN et al. 1987).
The first canonical axis ("x-axis") extracted by CCA was closely related to the management practices, as indicated by long vectors and nearby cen- troids of nominal factors (Fig. 5) and by high inter- set correlations with the axis (Table 4).Continuous herbicide application proved to be the most "effective" factor explaining the composition of weed flora.
The second axis ("y-axis") was associated with soil variables, particularly pH, and with climatic factors, precipitation and effective temperature sum between sowing and sampling.Galeopsis spp.and Polygonum spp.occurred frequently in moist or- ganic soils, whereas Sonchus spp., Poa annua and Lapsana communis thrived in coarse soils and warm and humid weather conditions which were typical of the eastern region of the survey.
Although the eigenvalues obtained by CCA were low, the first two canonical axes from the constrained ordination accounted for 49% of the total species-environment variation.In the analysis of weed density, the corresponding value was 53%.Partial CCA with regions as covariables slightly reduced the explained variance.The first canonical axis was statistically significant (P=o.ol,Monte Carlo permutation test) in all analyses.Fig. 3. Mean proportion (%) of the most abundant weed species of (a) the total weed density and (b) weed biomass in unsprayed spring cereal fields.Assessment was made in July.

Fig. 4. Ordination diagram based on redundancy analysis
(RDA) of weed densities describing indicator species for southwest (SW), central-east (CE) and central-west (CW)   regions of Finland.Centroids of all regions lie outside the range of the diagram.Some species near the origin are not shown because of their over- lapping position.
Explanatory factors: CER = Cereal-dominated rotation, CLAY = Clay soil, COARSE = Coarse soil, COVER = Crop cover, DRY = Dry soil, ETS = Effectice temperature sum between sowing and sampling, HERB = Duration of herbicide use, MIX = Mixed crop rotation, NORMAL = Normal soil moisture, ORGANIC = Organic soil, PH = Soil pHH 20, PREC = Precipitation sum between sowing and sampling, WET = Wet soil, YIELD = Crop yield.Fig. 5. Ordination diagram based on canonical correpondence analysis (CCA) of weed biomass data from 252 spring cereal fields.The end of dotted vectors lies outside the range of the diagram.Two species (STAPA, URTSS) near the origin are not shown because of their overlapping position with other species.Since the density values showed a skewed distribu- tion, the median value is a more appropriate meas- ure to indicate the level of weed infestation in spring cereal fields.The results correspond to the present weed infestation levels found in field ex- periments in the Nordic countries (Hallgren  1993, Salonen 1993).
The weed flora was dominated by rather few species (Table 1, Fig. 3) which is a common phenomenon in intensified farming systems (Neururer 1965, Callauch 1981, Albrecht and Bachthaler 1988).The low number of abundant species makes e.g. the choice of herbicides easier.The mean proportion of individual weed species out of the total weed density and biomass in each field (Fig. 3) indicated that Chenopodium album, Stellaria media and Viola arvensis are the most dominantspecies in terms of weed density whereas, Galeopsis spp., C. album and Elymus repens were the most dominant species in terms of biomass production.Furthermore, Stellaria media and Viola arvensis had a higherproportion in densities than in biomass.
The most aggressive weed species such as Galeopsis spp.and volunteer turnip rape (BRSRO)  were detected both by their proportion of total weed biomass (Fig. 3) and along the crop cover vector (COVER) in the ordination diagram (Fig. 5).
Differences in weed abundances and weed biomass production between soil types (Fig. 2) infer both to growth conditions and species composition.Apparently, differences in weed growth be- tween soil types are also reflected in yield re- sponses of the crop.Indeed, yield responses of cereals have been found to be the lowest in clay soils (Jensen 1985, Hallgren 1989).
Ordination analysis provided easily interpretable results which were in agreement with the conclu- sions based on the regression analysis (Erviö and  Salonen 1987) with regard to the most important factors affecting the occurrence of weeds.How- ever, the relative importance of different factors was more clearly and easily pointed out by the ordination analysis than by the regression analysis.A particular advantage in applying ordination ana- lyses is that the CCA ordination diagrams are not in any way hampered by high correlations between weed species or between environmental variables (Ter Braak 1987 c).
The most frequent species like Chenopodium album and Viola arvensis located near the origin of the ordination diagram (Fig. 5).These species were found in all field types indicating that they are well adapted to agricultural ecosystem.
Each geographical regions had its characteristic weed species as was also concluded earlier with the analysis of variance (Erviö and Salonen 1987).Some less frequent species like Achillea spp., and Galium spp.were particularly associated with cer- tain regions (Fig. 4).
The results achieved with RDA (Fig. 4) and CCA (Fig. 5) can be combined.Typical weed species for cereal-dominated rotations with frequent use of herbicides were Lamium spp., Galium spp., Fumaria officinalis, Tripleurospermum inodorum and volunteer turnip rape.In addition, Lamium spp.and Galium spp.thrived in clay soils as reported also by mukula et al. (1969) and Andreasen et al. (1991).Indeed, cereal-dominated crop rotations were common in south-western Finland (SW) where most of the fields were clay soils and herbi- cides were frequently used.In this region, turnip rape and winter cereals are common crops in rota- tion, thus promoting the occurrence of volunteer oilseed rape and Tripleurospermum inodorum which is a common species in wintercereals (Raa- tikainen et al. 1978).
Long-term use of herbicides has evidently se- lected the weed populations in south-western Finland towards the more tolerant species like Galium spp., Lamium spp.and Tripleurospermum ino- dorum.Indicator species for geographical regions hopefully help e.g.advisory services to direct general control recommendations to differentregions.Some weed species were typical of central Finland.Rumex spp. (R. acetosa and K. acetosella) and Ranunculus repens are associated with grassland (Raatikainen and Raatikainen 1975) which is a common crop in rotations in central Finland.Polygonum spp.were characteristically in organic soils which were most often found in central Finland.
Both management practices and soil properties affected the weed vegetation.However, the effects of individual environmental variables should be interpreted with caution due to the confounded na- ture of several factors associated with crop rotation and crop management.Similar conclusions were drawn from the regression analysis (Erviö and Sa- lonen 1987) and in other weed surveys (Streibig et al. 1984, Andreasen et al. 1991, Dale et al. 1992).
In general, the possibilities to draw definite con- clusions from this kind of weed survey data seem to be limited due to the complex cropping history in each field as discussed also by Cousens et al.  (1988).Relatively high eigenvalues of the third and fourth axes (Table 4) indicated the complex nature of weed communities which was impossible to ex- plain with the few environmental variables chosen.Obviously, more detailed information on the cropping history, like crop rotation, type of herbicides used and their control efficacy, is needed to de- scribe better the responses of weed flora to environmental conditions.Forward selection of environmental variables in CANOCO appeared to be rather liberal in judging the variables statistically significant.This is a typical shortcoming of stepwise selec- tion of variables because the overall size of the test is not controlled (Ter Braak 1990).
Low eigenvalues of ordination axes reveal either short environmental gradients in cereal fields or the plasticity of most weed species to grow in diverse conditions.Low eigenvalues are also the con- sequence of few abundant species which occurred in most fields.As the number of observed weeds was limited to 33 taxa, the study probably ignored some weed species thatmight have had an explanatory power in the ordination analysis.Nevertheless, ordination diagrams illustrated some clear patterns in weed communities and related environmental factors and agricultural practices to the weed data.Ordination techniques detected groups of weed species similar to those by the TWINSPAN classi- fication of the same data (Salonen 1990).
In conclusion, weed flora showed a response particularly to management practices.Herbicide use affected most the composition of weed flora.This should be considered in planning the crop rotations and long-term weed control measures.The ordination analysis detected also an evident relationship between the weed composition and soil properties in the field.Some of these properties, e.g.soil pH, can be manipulated if neccessary.
Simultaneous analysis of several weed species with environmental factors and plotting of data gave a better community level description on weed incidence than regression analysis (Erviö and SA- LONEN 1987) which rather provided a deeper but narrower insight into weed vegetation and individ- ual weed species.Still, the ordination analysis should be regarded both as a hypothesis provoking and explaining approach.Nevertheless, it is appropriate to start the analysis of data from weed sur- veys with ordination analysis and to continue, if necessary, with regression analysis to solve more detailed hypotheses.
are according to the BAYER standard(BAYER 1992).

Fig. 1 .
Fig.1.Distribution of spring cereal fields into weed infestation classes according to a) weed density and b) air-dry biomass Assessment was made from unsprayed sample plots in July.

Fig. 2 .
Fig. 2. Weed infestation in different soil types.The mean weed density (left bar) and air-dry biomass (right bar) in unsprayed fields.Vertical line indicates the standard error of the mean.

Table 2 .
Environmental variables subjected to the canonical correspondence analysis (CCA).

Table 3 .
Eigenvalues (X M )corresponding to the first four ordination axes from Correspondence Analysis (CA) and Canonical Correspondence Analysis (CCA).Environmental values for CCA are given in Table2.Partial analyses were performed with regions as covariables.Weed infestation values from 252 spring cereal fields were transformed with ln(y + I).

Table 4 .
Inter-set correlations of environmental variables with the first four ordination axes from CCA for the weed biomass data.The two highest values of each axis are underlined.