A 38-year study of relations between weather and vegetation dynamics in road verges near Bibury, Gloucestershire



1 The Bibury long-term data set contains information on annual fluctuations in the abundance of over 100 grasses and forbs in roadside verge vegetation over the period from 1958 to the present. Monitoring has been carried out every July by the same individual. The data set represents a unique long-term record of the dynamics of a complete plant community.

2 Records for the most abundant taxa (including bare ground and litter) were used to determine the effect of climate variability on the year-to-year performance of the selected species. Residuals about the long-term mean log biomass of each species (de-trended where the species showed a significant increase or decrease in abundance over time) were correlated against indices of interannual climate variability. Plant and weather records were compared over 3-month seasonal periods (March–May, June–August, September–November, December–February) or 6-month seasonal periods (March–August, September–February), with time lags of 0, 1 and 2 years.

3 Principal components analysis (PCA) was used to formulate annual weather indices, using either conventional weather variables (temperature, rainfall and sunshine) or the Lamb catalogue of daily weather types.

4 Between 5% and 70% more correlations were observed than might be expected to occur by chance, depending on the season and the PCA index, indicating markedly non-random plant–weather relationships. Total vegetation production was positively correlated with minimum spring temperature. The distribution of correlations was generally evenly distributed across the three lag periods.

5 In general, those species favoured by environmental stress or disturbance were promoted following warm dry springs and summers, whereas those favoured by more productive conditions were promoted following a wet growing season.


Annual monitoring of mesotrophic grassland vegetation in wide road verges near Bibury, Gloucestershire, started by A. J. Willis and E. W. Yemm in 1958, has resulted in a unique record of the performance of individual species within the verges, and of the vegetation as a whole, over a very long period. Results from the first 30 years of the study, concerning the management of roadside vegetation through the application of herbicides and a growth regulator, have been widely quoted and applied (Willis 1972; Willis 1988). However, in the late 1980s, as interest increased in the possible effects of climate change on the functioning of ecosystems, the Bibury records had a new lease of life with the full realization that data from the untreated control plots could yield valuable insights into the relations between plant performance and climate (Grime et al. 1994).

Despite the increasing number of long-term studies now being published (e.g. Dodd et al. 1995; Fitter et al. 1995; Sparks & Carey 1995) the Bibury data set has several features that make it unique. The information has been collected by a single recorder over the entire period, using a consistent sampling method. The records are continuous and contain information on the long-term dynamics of a complete plant community.

The aim of this study was to investigate the influence of weather on annual fluctuations in plant performance using the Bibury data set. To achieve this, annual vegetation records were compared with time series both of basic meteorological data (individual weather variables) and of higher-level meteorological data (synoptic patterns). We determined whether plant performance was related to weather in the various seasons and over a range of time lags between occurrence and observation. Observed relations are discussed not only in terms of individual species’ responses but also with regard to functional types and plant life strategies.

Site description

The vegetation records were taken from permanent plots set up along a 700-m stretch of Akeman Street, originally a Roman road, near Bibury, Gloucestershire (National Grid reference SP 119048). The site was initially chosen for the exceptional width of its road verges, the apparent homogeneity of its vegetation, and because of the fairly light traffic along this length of minor road. The original experiments were established to monitor the effects of spraying the herbicide 2,4-D and the plant growth regulator maleic hydrazide onto the vegetation of road verges. The controls for these experiments consisted of eight plots of unsprayed vegetation, each 20 m long and 3.3–5.5 m wide. Data from these untreated control plots were used in the analysis described below to investigate plant–weather relationships. The plots can be divided into two groups, plots 1–6, which form controls to the maleic hydrazide treatments, with and without 2,4-D, and plots 7 and 8, which are at the western end of the experimental site in what appears to be a more ancient part of the verge and which form controls to the 2,4-D-only treatments. Each plot consists of a rectangular area within which is set a permanent quadrat of 1.0 × 0.5 m, in most cases about 2.0 m from the roadside edge of the plot.


The parent geological stratum of the Bibury area is primarily oolitic limestone. The depth of the soil generally varies between 300 mm and 800 mm, although in places it can be as little as 100 mm, and pH values lie within the range 7.5–7.9. Pollution and contamination from vehicle emissions fall off rapidly with distance from the road edge. In a study of the verges of roads with very heavy traffic (chiefly A roads), Akbar (1997) found that contamination and pollution were sharply reduced with distance from the road, with concentrations of most heavy metal elements only significantly above normal within the border zone of a road (up to 1 m from the edge.) The permanent quadrats at Bibury, which are all situated well-back from the road, which is subject to only light traffic, are therefore not likely to be affected.


The vegetation of the Bibury road verges can be characterized under the National Vegetation Classification (Rodwell 1992) as an Arrhenatheretum elatioris grassland (MG1). Two grass species, Arrhenatherum elatius (nomenclature follows Stace 1997) and Dactylis glomerata, are the major components of the vegetation and comprise 40% of the total above-ground biomass. The large Umbellifers Anthriscus sylvestris and Heracleum sphondylium are the most conspicuous herbs. Well over a hundred taxa have been recorded from the Bibury road verges but only the most abundant taxa (listed in Table 1), including bare ground and litter, were included in this analysis.

Table 1.  Long-term trends, the presence of autocorrelation and a functional classification of major taxa in the Bibury data set. Taxa in plots 1–6 are represented by A and taxa in plots 7 and 8 are represented by B

Taxa exhibiting long-term increase in abundanceTaxa exhibiting long-term decrease in abundanceTaxa that do not exhibit autocorrelationFunctional type sensuGrime 1979 (from Hunt et al. 1993)
Litter   B
Bare ground   A B
Achillea millefolium    CR
Agrostis stolonifera    CR
Alopecurus pratensis (1–6 only)   A CSR
Anthriscus sylvestris   B CR
Arrhenatherum elatiusA   C
Brachypodium pinnatum (7 and 8 only)  B B SC
Bromopsis erecta   B SC
Centaurea nigra (7 and 8 only)B   S
Cirsium arvense  A B CR
Convolvulus arvensis  A  CR
Cruciata laevipes  A A B CSR
Dactylis glomerataB   C
Elytrigia repens  A B   CR
Festuca arundinacea (1–6 only)A  B CSR
Festuca rubra  A B CSR
Galium aparine   B CR
Galium verumA   SC
Glechoma hederacea  A B  CSR
Heracleum sphondylium   B CR
Hypericum maculatum (7 and 8 only)  B  CSR
Knautia arvensisA B  A CSR
Lolium perenneB  A CR
Odontites vernusA B  A B R
Phleum bertolonii (1–6 only)A  A B CSR
Plantago lanceolataB   CSR
Poa pratensis  A A B CSR
Potentilla reptans (7 and 8 only)B  B CR
Ranunculus repensA B   CR
Rumex crispusB  A B CSR
Taraxacum officinale agg.B  B CSR
Tragopogon pratensis (7 and 8 only)B  B CR
Trifolium pratense   A B CSR
Trifolium repens    SC
Trisetum flavescens   A CSR
Urtica dioica  A B  C
Veronica chamaedrys   B S
Vicia sativaA

Although the roadside verges were chosen as an experimental site partly because of the uniformity of their vegetation, some variation is present. A distinction can be made between plots 1–6 and plots 7 and 8. The vegetation in plots 1–6 tends towards the more vigorous, less diverse Festuca rubra and Urtica dioica subcommunities (MG1a and MG1b, respectively), while the vegetation in plots 7 and 8 is more typical of the Centaurea nigra subcommunity (MG1f). A fuller account of the floristic composition of the verges is given by Yemm & Willis (1962).

A metre-wide strip of the verges adjacent to the road is mown in late spring to maintain visibility, and the rest of the verge is usually ‘topped’ with a flail mower in November each year at a height of approximately 0.5 m.

Over the long-term monitoring period there has been no decrease in species-richness in plots 1–6, while in plots 7 and 8 there has been a significant increase in the number of species recorded (r = 0.432). The abundance of different species has fluctuated considerably. Regressions of log biomass against time indicate that 16 species have shown a significant increase in abundance, 10 a significant decrease, while 12 have shown no overall change (Table 1). Of those species which have significantly increased in abundance, such as the annuals Odontites vernus and Vicia sativa and the perennials Lolium perenne, Ranunculus repens, Rumex crispus and Taraxacum officinale agg., a number are advantaged by some disturbance, while others, such as Arrhenatherum elatius, Centaurea nigra, Festuca arundinacea, Galium verum and Knautia arvensis, exhibit a degree of drought tolerance. Species that have shown a significant decrease in abundance include a number that can tolerate some shade, such as Cruciata laevipes and Glechoma hederacea, or which are not typical of dry grassland, such as Poa pratensis and Urtica dioica. These trends may reflect the long-term trends in summer rainfall discussed below and shown in Fig. 1.

Figure 1.

Mean monthly summer rainfall (mm) over the period 1956–96. r = −0.382, P < 0.05.

Materials and methods


Vegetation records are made annually in the third week of July. The non-destructive recording method used is provided in full by Willis et al. (1959) and Yemm & Willis (1962). The permanent quadrat in each plot is marked by wooden posts, within which is fitted a wire grid (1.0 × 0.5 m) divided into eight 250 × 250 mm subsections. Each subsection is scored separately by subjective estimation of ‘relative bulk’. Each species occurring in that subsection is given a score out of 10 (after allowances have been made for bare ground and litter), according to the proportion of the total above-ground biomass that the species constitutes. Species making up less than 5% of the total bulk of the vegetation are scored as ‘traces’, and in calculating average results all traces are arbitrarily assessed as 1%.

In order to obtain comparative estimates of the total above-ground biomass of the vegetation of the plots, a subjective estimate of the volume of living plant material in each of the subsections is made by reference to a standard volume (a cylinder of 10 mm diameter and 150 mm height). The mean of the eight scores for the subsections is calculated to give a whole quadrat score.

The estimates of vegetation bulk have been compared with those of destructive samples, for which the fresh mass and the dry mass were also determined, taken periodically throughout the long-term study. No significant differences (P < 0.05) were found between the destructive and non-destructive estimates of biomass or of relative abundance.


The field scores were converted to biomass values by means of calibrations derived periodically from small destructive samples of mixed vegetation from the control plots. Total field bulk scores were converted to biomass values using the formula: one biomass unit (field estimate) = 5.9 g (dry mass) of plant material, thus giving a mean value for the different types of vegetation present. Biomass values for each species were calculated as percentages of the total vegetation biomass. The data were therefore in the form of biomass per unit area. All biomass values were standardized to g m−2 and transformed to logarithms before further analysis. Field estimates of biomass were made from 1959 onwards, therefore the transformed vegetation time series run for the 38-year period between 1959 and 1996.


Several of the species showed significant trends in mean log shoot biomass over time, and to counter any influence of long-term environmental trends this non-stationarity was removed statistically before climatic correlations were attempted. Data for those taxa that did show long-term increases or decreases in log biomass (Table 1) were ‘de-trended’ by calculating the residuals (observed values minus fitted values) about the fitted 38-year linear regression of log biomass on time. The plant data were then replaced by those residuals. Those species that showed no significant increase or decrease in log biomass over time were ‘flat-trended’. This was done by calculating the residuals about the 38-year mean of log biomass for each individual species, with residuals replacing the original data, as above.


Taxa were classified into functional groups according to their established plant strategies within the C-S-R system (sensuGrime 1979). A functional classification of components of the Bibury vegetation is shown in Table 1: representatives of all groups apart from stress-tolerant ruderals are present. The total log biomass for each functional type was calculated and de-trended or flat-trended as above. In addition, the total vegetation biomass was flat-trended.


Principal components analysis (PCA) was used to detect relationships between intercorrelated weather variables and to formulate indices that characterized the weather in any given season. The method identifies statistically independent linear combinations of variables that account for decreasing amounts of the total variance in a data set (Jones & Kelly 1982). 3-monthly periods were used: autumn (September–November), winter (December–February), spring (March–May) and summer (June–August), and 6-monthly means were then calculated for autumn plus winter and spring plus summer. Because vegetation records are collected at the end of July each year, compatability when conducting plant–weather correlations was obtained by calculating summer means for the current year from June and July data. Fixed seasonal periods were chosen to allow comparisons to be made of the response of different species to weather in the same season.

PCA was undertaken for each 3- and 6-month combination, using minitab (Release 9 1993). The contribution of each weather variable is given by the loading of that variable on the component. A series of amplitudes gives the strength of the component from year to year. Only the results for the first component (accounting the greatest variance) are discussed here.


Weather variables were obtained from RAF Lyneham (a representative station within 25 km of the experimental site). Four weather variables were used in this analysis: monthly means of daily maximum temperature and daily minimum temperature, monthly total rainfall and monthly total sunshine. Seasonal 3- and 6-monthly means of the monthly means were calculated. Regression against time showed that there had been no significant trend in temperature or sunshine in any season over the 38-year floristic recording period, but that there had been a significant decrease (Fig. 1) in summer rainfall. Over that period, drought years have become progressively drier, and mean monthly rainfall in very wet summers has progressively declined. All variables apart from summer rainfall were flat-trended by calculating the residuals around their long-term mean over the period 1956–96. The summer rainfall variable was de-trended by calculating the residuals about the fitted 40-year linear regression of rainfall on time (the 40-year period enabled comparisons to be made between weather and plant performance with time lags of up to 2 years).

The loadings of the weather variables on the first component and the percentage variance for each season are shown in Table 2a.

Table 2.  The loadings of the first component and the percentage variance accounted for by the first component for each season

AutumnWinterSpringSummerAutumn and winterSpring and summer
  1. **P < 0.01, ***P < 0.001.

(a) Conventional weather type index
% variance loading48.454.
Minimum temperature−0.32−0.65 0.12 0.49 0.68 0.35
Maximum temperature−0.65−0.65 0.56 0.55 0.66 0.60
Rain 0.43−0.34−0.55−0.47 0.31−0.48
Sun−0.54 0.20 0.61 0.49−0.04 0.54
(b) Lamb daily weather type index
% variance loading5353.854.555.550.358.1
Anti-cyclonic−0.575 0.598 0.743 0.711 0.739 0.588
Cyclonic 0.736−0.734−0.54−0.670−0.574−0.690
Westerly−0.357 0.321−0.39−0.212−0.353 0.422
(c) Correlation between the two indices
r 0.478** 0.086 0.644*** 0.647***−0.105 0.618***

Both spring and summer, and the 6-month spring and summer combined, show similar patterns, with positive loadings on temperature and sunshine and negative loadings on rainfall. A positive amplitude therefore indicates a season with an increased frequency of relatively dry, warm and sunny weather. For autumn, temperature and sunshine have negative loadings, while rainfall has a positive loading. A positive amplitude therefore indicates an increased frequency of wet, cool, cloudy weather. In winter, temperature and rainfall have negative loadings, while sunshine has a positive loading. A positive amplitude therefore indicates an increased frequency of sunny, cold, dry weather. The combined 6-month index for autumn and winter reflects the winter pattern. In all cases the index derived from the first component can be clearly related to the frequency of rainfall and its associated weather conditions (increased rainfall being associated with relatively cool temperatures and increased cloud in spring, summer and autumn, and relatively less cold temperatures and increased cloud in winter).


While the relative frequencies of individual weather variables can be used to typify the weather in any individual season, much of the character of the British weather and climate is related to the direction, nature and persistence of the wind direction. Different wind directions produce different types of weather, which vary in character according to the season (Musk 1988). This association between wind direction and definable meteorological conditions forms the basis of the Lamb Daily Weather Type Catalogue (Lamb 1972). The Lamb system classifies the weather of Britain into different ‘weather types’ and has been used extensively in studies of the climate of the British Isles (Jones & Kelly 1982). A weather type is a definable entity that often lasts for several days whilst the weather undergoes variations typical of the succession of air masses and depression tracks occurring with that prevailing wind direction and general type of weather sequence or spell (Lamb 1964). Classifying weather according to weather type can be seen as a ‘higher order’ classification than that involving the use of individual weather variables alone.

A total of 27 weather types has been identified and these are usually condensed into seven major weather types: five are directional (westerly, north-westerly, northerly, easterly and southerly) and two are used when a synoptic system dominates the region (anti-cyclonic and cyclonic). Hybrids between the different types are possible and are recognized when two or three of the types defined above are combined. Examination of daily synoptic charts for the British Isles enables individual days to be classified according to these weather types.

A study of the frequency of the different weather types over the hundred-year period 1871–1971 showed that the anti-cyclonic and westerly types occurred on approximately 50% of all days, while 2 days out of 3 were either westerly, cyclonic or anti-cyclonic (Lamb 1972). Because these three weather types largely define the character of the weather of the UK, they were selected for the analysis described in this paper. Descriptions of the three types are given by Lamb (1964).

Lamb daily weather type (LDWT) data were obtained from the Climate Research Unit, University of East Anglia, for every day from January 1956 to May 1996. Monthly totals were calculated from these data for each of the three major weather types, using Lamb's standard accounting procedures (Jones & Kelly 1982). Seasonal mean frequencies were then compiled. The meteorological data were flat-trended and the residuals around their long-term means were calculated.

PCA was undertaken for each season. The loadings of the LDWT variables on the first component and the percentage variance for each season are shown in Table 2b.

There is a difference in the relative loadings of the three components between spring and summer, and autumn and winter. A positive amplitude of the first principal component in spring and summer, and in the 6-month spring and summer combination, indicates a season with relatively high frequencies of anti-cyclonic days and low frequencies of cyclonic and westerly days. This clearly equates with a dry, warm season. A positive amplitude in winter, and the 6-month autumn and winter combination, is associated with relatively high frequencies of anti-cyclonic and westerly days, and a low frequency of cyclonic days. This represents seasons characterized by periods with a generally westerly synoptic situation punctuated by anti-cyclonic episodes, which are likely to be relatively bright and dry, as opposed to wet, mild seasons dominated by cyclonic conditions. The loadings in autumn show the same relationship, but with opposite signs.

To determine the similarities, if any, between the PCA-derived indices derived from LDWT and conventional weather variables, the residuals of the weather-type index were correlated against the residuals of the weather variable index in each 3- and 6-month period. The correlation coefficients derived are shown in Table 2c.

There are clear similarities between the two indices in spring, summer and autumn, and in the combined spring and summer period. The similarity between the two indices is particularly strong in the spring and summer periods and indicates that positive values for each index can be equated with relatively dry, warm and sunny weather. Again in autumn there is a clear similarity: positive values for both indices are equated with mild, cloudy and wet weather. Comparisons can not be made between the indices in winter and over the combined autumn and winter period.


The log biomass vegetation scores were correlated with the seasonal meteorological indices. Correlations were made between plant performance and weather with lags of 0, 1 and 2 years. The number of observed significant correlation coefficients was compared with those expected to occur by chance. Thirty-four of the 40 selected taxa occur in plots 1–6 and 36 in plots 7 and 8; correlations were made separately for both sets of plots, even for species that occur in both. Therefore the number of expected correlations (at P < 0.05) would be (34 × 12) + (36 × 12) × 0.05 = 42 for the 3-month seasonal periods with lags of 0, 1 and 2 years, and (34 × 6) + (36 × 6) × 0.05 = 21 for the 6-month seasonal periods.


Because of the predominantly perennial nature of the vegetation, the time series for each Bibury species was examined for the presence of autocorrelation using the Durbin–Watson test (Wigley et al. 1985). Approximately 75% of all taxa in plots 1–6 exhibited statistically significant autocorrelation, and approximately 50% in plots 7 and 8. In plots 1–6 in particular, the dominant species tend to exhibit autocorrelation. The difference between the two series may be a result of the greater diversity of plots 7 and 8, and a lesser degree of dominance of the vegetation by a few vigorous species. Those species that did not exhibit statistically significant autocorrelation are indicated in Table 1. Where autocorrelation was found to be statistically significant, the number of degrees of freedom that could be used for testing correlations between plant performance and weather was reduced using the formula of Quenouille (Hays et al. 1993), which calculates the effective number of independent observations (E):

E = N/(1 + 2r1r1 + 2r2r2)

where N is the number of points in each of the two series to be correlated, r1 and r1 are the lag-one autocorrelations of the two series and r2 and r2 are the lag-two autocorrelations of the two series. The significance of correlations between plant biomass and meteorological variables was then estimated using Fisher's Z-test (Freund 1988).



The full results of the comparison between plant performance and conventional weather variables (CWV) are shown in Table 3. Table 4 lists the numbers of species showing negative or positive responses to the PCA index in the different seasons, and with different lags. More taxa exhibited a response in spring than any other 3-month season (Table 4). In total, 50 significant correlation coefficients were observed for the 3-month seasonal periods (Table 3), an increase of 20% over that expected to occur by chance alone. Over the 6-month seasonal period, 28 significant correlations were observed, an increase of 30% over expectation. More correlations were observed for year 0 than for lags of 1 or 2 years (Table 4).

Table 3.  Comparison of performance of taxa with the conventional weather variable PCA index for the 3- and 6-month seasonal periods with lags of 0, 1 and 2 years. Correlation coefficients are significant at P < 0.05. Figures in bold are from plots 1–6, others are from plots 7 and 8. Column headings refer to 3- or 6-month seasonal periods, with lags of 2, 1 or 0 years

Au2Wi2Sp2Su2Au1Wi1Sp1Su1Au0Wi0Sp0Su0Au & Wi2Sp & Su2Au & Wi1Sp & Su1Au & Wi0Sp & Su0
Litter   0.412               0.408
    0.354               0.397
Bare ground   0.403     −0.419−0.363   0.337    0.384 
Achillea millefolium     0.370             
Agrostis stolonifera   0.421               
Alopecurus pratensis       0.332           
Anthriscus sylvestris            0.460      0.399
Arrhenatherum elatius
Brachypodium pinnatum 0.469       0.369          
Bromopsis erecta      0.354    0.346      −0.339−0.355
Centaurea nigra
Cirsium arvense0.353       0.416       0.363  
Convolvulus arvensis
Cruciata laevipes  0.3970.436        0.358      0.390
Dactylis glomerata   −0.510        −0.366−0.394    
Elytrigia repens  0.417          0.357 0.389     0.393
Festuca arundinacea
Festuca rubra     0.410        −0.340   
Galium aparine                 0.349
Galium verum      0.548         −0.351 
Glechoma hederacea  0.428          −0.439 −0.405  0.360 
Heracleum sphondylium            −0.473     
Hypericum maculatum         0.424 −0.444       
Knautia arvensis
Lolium perenne   0.359 −0.333     −0.387       
Odontites vernus       0.584           
Phleum bertolonii     −0.357 0.544 0.368        0.537  
Plantago lanceolata
Poa pratensis  −0.362               
Potentilla reptans   0.372               
Ranunculus repens   −0.389              
Rumex crispus −0.374   −0.349 0.417       0.372 0.405   
Taraxacum officinale agg.−0.377  0.580    0.502         0.462  
Tragopogon pratensis
Trifolium pratense   0.467               
Trifolium repens    −0.366             
Trisetum flavescens        0.415        0.454  
Urtica dioica
Veronica chamaedrys−0.442−0.353 0.476          0.512     
Vicia sativa −0.356 0.422          0.423
Table 4.  Numbers of correlations between plant biomass and conventional weather variables in the 3- and 6-month seasonal periods, and with lags of 0, 1 and 2 years. Lag numbers in parentheses refer to the 6-month periods; lag numbers not in parentheses refer to the 3-month periods

Response to the PCA indexAutWinSprSumAu and WiSp and SuLag 0Lag 1Lag 2


Total810179141224 (9)17 (8)9 (7)

Examples of species showing positive and negative relationships with the spring PCA index are shown in Fig. 2. The results suggest that Phleum bertolonii and Taraxacum officinale agg. are either promoted following warm dry springs or retarded following cool wet springs, while Galium verum and Hypericum perforatum clearly exhibit the opposite responses.

Figure 2.

Examples of positive and negative relationships with the PCA index for conventional weather types. (a) Phleum bertolonii spring, t-1, r = 0.544, plots 1–6. (b) Galium verum spring, t-1, r = –0.548, plots 7 and 8. (c) Taraxacum officinale agg. spring, t-1, r = 0.502, plots 1–6. (d) Hypericum maculatum spring, t-0, r = – 0.44, plots 7 and 8. All residuals have been standardized (zero mean, unit variance).


The full results of the comparison between plant performance and the Lamb weather type index are shown in Table 5. Table 6 lists the numbers of taxa showing negative or positive responses to the PCA index in the different seasons. In total, 44 significant correlation coefficients were observed for the 3-month seasonal periods. This is an increase of only 5% over the number expected to occur by chance. However, over the 6-month seasonal periods, 36 significant correlations were observed, an increase of 71% over expectation. There was an even distribution of correlations across years (Table 6).

Table 5.  Comparison of performance of taxa with the Lamb daily weather type PCA index for the 3- and 6-month seasonal periods with lags of 0, 1 and 2 years. Correlation coefficients are significant at P < 0.05. Figures in bold are from plots 1–6, others are from plots 7 and 8. Column headings refer to 3- or 6-month seasonal periods, with lags of 2, 1 or 0 years

Au2Wi2Sp2Su2Au1Wi1Sp1Su1Au0Wi0Sp0Su0Au & Wi2Sp & Su2Au & Wi1Sp & Su1Au & Wi0Sp & Su0
Bare ground




Achillea millefolium


Agrostis stolonifera
Alopecurus pratensis

Anthriscus sylvestris0.342



Arrhenatherum elatius





Brachypodium pinnatum0.441

Bromopsis erecta



Centaurea nigra0.446

Cirsium arvense


Convolvulus arvensis


Cruciata laevipes



Dactylis glomerata



Elytrigia repens


Festuca arundinacea


Festuca rubra
Galium aparine
Galium verum


Glechoma hederacea


Heracleum sphondylium




Hypericum maculatum
Knautia arvensis

Lolium perenne
Odontites vernus


Phleum bertolonii
Plantago lanceolata


Poa pratensis


Potentilla reptans


−0.362 0.401
Ranunculus repens


 0.437 0.405



Rumex crispus



Taraxacum officinale agg.





Tragopogon pratensis


Trifolium pratense


Trifolium repens

Trisetum flavescens





Urtica dioica
Veronica chamaedrys



Vicia sativa



Table 6.  Numbers of correlations between plant biomass and Lamb weather types in the 3- and 6-month seasonal periods, and with lags of 0, 1 and 2 years. Lag numbers in parentheses refer to the 6-month periods; lag numbers not in parentheses refer to the 3-month periods

Response to the PCA indexAutWinSprSumAu and WiSp and SuLag 0Lag 1Lag 2


Total914712141617 (14)13 (10)15 (12)

Examples of positive and negative relationships with the PCA index are shown in Fig. 3. The results suggest that Veronica chamaedrys is either promoted following relatively dry bright winters or is retarded following cloudy dull winters, as is Potentilla reptans, while Arrhenatherum elatius shows the opposite response. Dactylis glomerata is either retarded by warm dry springs and summers or promoted by cool wet springs and summers.

Figure 3.

Examples of positive and negative relationships with the PCA index for Lamb daily weather types. (a) Veronica chamaedrys winter, t-2, r = 0.600, plots 7 and 8. (b) Dactylis glomerata spring and summer, t-2, r = –0.705, plots 1–6. (c) Potentilla reptans autumn and winter, t-0, r = 0.401, plots 7 and 8. (d) Arrhenatherum elatius winter, t-1, r = –0.459, plots 7 and 8. All residuals have been standardized (zero mean, unit variance).


The total above-ground biomass of the vegetation was correlated against the seasonal PCA indices, and in addition with the individual weather variables (maximum, minimum and mean temperatures, sunshine and rainfall) for the 3- and 6-month periods. Biomass was found to be positively correlated at P < 0.05 only with minimum temperature in the current spring, i.e. in the period March–May immediately preceding vegetation recording, in both plots 1–6 and 7 and 8 (r = 0.397 and 0.497, respectively), and with mean temperature in the current spring in plots 7 and 8 (r = 0.436). Biomass was also found to be negatively correlated with the autumn and winter combined weather variable index with a lag of t-2, and with the spring and summer combined weather type index, again with a lag of t-2.


The full results of the comparison between the performance of functional groups and seasonal fluctuations in weather are shown in Table 7.

Table 7.  Comparison of the performance of functional types with the CWV index and the LDWT index for the 3- and 6-month seasonal periods with lags of 0, 1 and 2 years. Correlation coefficients are significant at P < 0.05. Figures in bold are from plots 1–6, others are from plots 7 and 8. Column headings refer to 3- or 6-month seasonal periods, with lags of 2, 1 or 0 years

Au2Wi2Sp2Su2Au1Wi1Sp1Su1Au0Wi0Sp0Su0Au & Wi2Sp & Su2Au & Wi1Sp & Su1Au0 & Wi0Sp & Su0
Conventional weather variables






Lamb daily weather types














More correlations were observed between the interannual abundance of the different taxa with interannual weather variation than are expected to occur by chance alone, strongly suggesting that non-random relations exist between plant and weather variables. The use of PCA-derived indices allows objective classification of plant response to seasonal weather conditions in a way that is not possible when using individual weather variables with a large vegetation data set.

The use of the two indices (CWV and LDWT) enables a more comprehensive picture to be built up of plant–weather relationships in the Bibury road verges than the use of each index alone would. This study extends the previously very limited application of LDWT to biological data. Masterman et al. (1996) used LDWT to model the autumn migration of the bird cherry aphid, Rhopalosiphum padi.Aebisher et al. (1988) compared the relationship between abundance of organisms at different trophic levels in the Atlantic Ocean over the period 1958–88 with the annual frequency of westerly weather: striking similarities were apparent between the frequency of westerly weather and biological abundance. We have demonstrated that LDWT can also be applied to terrestrial vegetation dynamics. Overall, nine taxa exhibited comparable responses to both indices in the same season: bare ground, Achillea millefolium, Brachypodium pinnatum, Bromopsis erecta, Dactylis glomerata, Elytrigia repens, Taraxacum officinale agg., Veronica chamaedrys and Vicia sativa. Of these correlations (11 in all), for all but two, the significance of the correlations with the LDWT index was greater than with the CWV index. The CWV index characterizes seasons according to the relative frequencies of individual weather variables, while the LDWT index relates these to prevailing synoptic conditions. As such, LDWT have potential use in predictive studies of the future effects of climate changes on vegetation performance. More research is needed to investigate the relationships between LDWT and other terrestrial data sets. Meanwhile, this study suggests that they give added value when used in conjunction with CWV.


A number of other workers have demonstrated that the total biomass production of grassland systems is linked to rainfall, either in the current season or the previous season, for example in English hay meadows (Smith 1960), and the Park Grass Experiments (Silvertown 1980; Silvertown et al. 1994). The latter authors found a positive relation with rainfall in the growing period prior to hay cut in the Park Grass plots. However, when the annual total above-ground production at Bibury was correlated with annual fluctuations in individual weather variables, rainfall was not found to be a significant controlling factor. Instead, mean spring minimum temperature was found to be the most important variable. The Bibury vegetation differs from the other examples given above in that it is not cut for hay but is instead lightly ‘topped’ at the end of the growing season. It may therefore be less dependent on summer rainfall to boost regrowth after cutting. It is likely that minimum temperatures in spring dictate the relative growth rates of the vigorous competitive species that dominate the Bibury verges, given that in most years moisture reserves at depth are likely to remain sufficient to support plant growth in the early part of the growing season, and that many of the most abundant species have an early phenology. The negative correlations with the PCA indices for autumn and winter combined and for spring and summer combined, both with a lag of t-2, are likely to reflect the performance of dominant species, such as Dactylis glomerata, in the sward.


A number of similarities are apparent in the responses of individual taxa to fluctuations in the PCA index for both weather types and conventional weather variables for the 3-month spring and summer periods, and for the 6-month spring and summer combined period. This is as expected, given the similar climatic characteristics of positive and negative amplitudes of both indices over these periods. The amount of bare ground in the system is promoted following positive years for both indices in spring. Similarly the amount of litter in the system is increased following warm dry springs and summers. The amount of litter would be expected to increase following a dry season, and this is not incompatible with an increased frequency of gaps at ground level in tall productive vegetation such as that in the Bibury verges. Species promoted by dry springs and summers include fairly deep-rooted species such as Centaurea nigra, Convolvulus arvensis, Taraxacum officinale agg. (illustrated in Fig. 2), Trifolium pratense and Rumex crispus. Other plants promoted include the vigorous drought-tolerant grasses, Arrhenatherum elatius, Brachypodium pinnatum, Elytrigia repens and Festuca arundinacea, and the annuals Odontites vernus and Vicia sativa. Species that are retarded by dry springs and summers include grasses typical of productive sites, such as Dactylis glomerata, Lolium perenne and Poa pratensis. Other species that are retarded are associated with shady or damp grassland and include Cruciata laevipes, Galium aparine, Glechoma hederacea and Ranunculus repens. Bromopsis erecta is also retarded by dry springs and summers at Bibury. This observation supports results from European populations of B. erecta, which are favoured by moist springs and retarded by a dry season (Bornkamm 1961). There are some exceptions to this general rule. For example, Galium verum, a species of well-drained calcareous soils, would be expected to be favoured by dry spring and summer conditions but it is in fact retarded.

Species that are promoted following mild or wet autumns and winters include those of damp or shady grassland, such as Alopecurus pratensis, Cruciata laevipes, Glechoma hederacea and Hypericum maculatum. Another example is Veronica chamaedrys, a perennial herb of moist grassland, which can form an understorey in tall grassland. It is promoted by wet autumns and winters at Bibury. This promotion may be enhanced by the species’ phenology: rapid shoot growth is restricted to spring and autumn (Grime et al. 1988). Bromopsis erecta is also promoted by mild winters and this may also be linked to phenology, it being a grass of early phenology, capable of growth during winter (Speddings & Dickmans 1972). Species promoted by cool dry autumns and winters include a number that are also favoured by dry springs and summers at Bibury, such as Rumex crispus, Taraxacum officinale agg. and Trisetum flavescens.


It appears that species’ life histories and habitat distributions can give some indication of their response to fluctuations in weather in mixed vegetation such as that in the Bibury verges. This is supported by the responses of the aggregated functional groups shown in Table 7. Warm, dry springs and summers promote those species that are adapted to environmental stress or disturbance [competitive ruderals (CR), ruderals (R) and stress-tolerators (S)], while species adapted to more productive conditions are retarded [competitors (C) and stress-tolerant competitors (CS)]. If, as predicted under climate change scenarios (e.g. CCIRG 1991), such summers become more frequent in southern Britain, the relative abundance of the more ruderal components of the Bibury vegetation might increase (Fig. 4). Those groups partially linked to the competitive strategy respond positively to mild wet winters, which, in effect, extend the productive growing season. The future balance between wet winters and dry summers will therefore be decisive in determining the relative abundance of components of the Bibury vegetation (Hunt et al. 1993).

Figure 4.

Functional groups retarded and promoted following warm dry springs and summers. The two environmental dimensions are: increasing stress, from top left to bottom right; increasing disturbance, from top right to bottom left. Those groups more tolerant of environmental stress or disturbance are promoted following such seasons, while competitors are promoted following cooler, wetter summers. After Hunt & Cornelissen (1997).

Interactions between plant performance and weather are complex: different weather variables may differentially affect flowering, seed production and vegetative growth. Plant responses may be direct or mediated through competition, and responses may lag behind weather conditions by one or more growing seasons. A striking feature of the results from Bibury is that the total number of correlations between plant performance and weather with a lag of 1 or 2 years is greater than that which occurs with weather in the current season, although overall there is a relatively even distribution of correlations over the three lag periods (0, −1 and −2 years). Time lags of more than 1 year between cause and effect are common features of natural systems (Magnuson 1990). Herben et al. (1995), using data from permanent grassland plots over a period of 10 years, also found large year-to-year variation in species’ performance that could be correlated with weather variables, and which also involved lagged responses. They noted that different measures of plant performance (biomass of ‘modules’ and number of modules) resulted in correlations with different variables over different periods for the same species, and suggested that this may be the mechanism by which time lags arise.

An example of the complexities involved in interpreting lagged responses is the relation between Dactylis glomerata and warm, dry springs and summers (Fig. 3). The same response (a negative relationship with a 2-year lag) is found with both PCA indices and in both vegetation series. Experiments with D. glomerata in mixed communities have shown that following productive spring conditions (warm and moist) the grass produces lush vegetative growth at the expense of inflorescences, and that the resulting large tussocks cause weakening of the less dominant members of the community; this effect is carried over into subsequent years (Dunnett 1996). The sensitivity to spring and summer drought may be linked to carbohydrate metabolism: the carbohydrate reserves of northern populations of D. glomerata (Volaire 1995; Volaire & Gandoin 1996) under long and intense drought are continuously utilized and not replenished, causing mortality and lack of persistence. Such effects may not be apparent in the field until seasons following the drought event. It may, of course, be that the observed effects are not direct responses but are mediated through interspecific competition. It is probable that many of the lagged responses of minor components of the Bibury vegetation represent reactions to the performance of dominants such as D. glomerata.

Field results obtained from long-term monitoring studies, such as those obtained from the Bibury verges, do not allow distinctions to be made between direct plant responses to weather variables and those mediated through interspecific competition. It has been suggested that positive feedback, favouring better competitors, may result in a magnification of the effects of climatic perturbations in fertile, productive habitats (Silvertown et al. 1994). In a series of microcosm experiments with synthesized grassland communities using transplants from the Bibury verges, we have shown that this is indeed the case. Asymmetric interspecific competition resulted in amplification of climatic signals, with more productive species being disproportionately promoted by favourable growing conditions (N. P. Dunnett & J. P. Grime, unpublished data).

The long-term record from the Bibury verges shows that even within what may be thought of as relatively stable vegetation there can be marked interannual fluctuations in species’ abundance. At least part of this variation can be explained as a response to climatic variability. Many of the observed responses match what is known of the species concerned through their natural history and published accounts. Several of the most frequent Bibury species have been subjected to climate manipulations under controlled conditions and their responses are the same as those observed in the field (Dunnett 1996). The Bibury data set therefore has significant potential in modelling the response of individual taxa, functional groups and the vegetation as a whole to any future climate scenario.


This work was supported by a studentship from the Esmée Fairbairn Charitable Trust. The late Professor E. W. Yemm collaborated in the recording for many years. We are grateful to Dr J. S. Conway, Royal Agricultural College, Cirencester, who carried out analysis of soils from the Bibury plots, and to H. Thomas and F. Volaire (IGER, Aberystwyth) for helpful comments.

Received 24 July 1997revision accepted 8 January 1998