TEXTUALIDAD E INTERTEXTUALIDAD DEL ÍTEM

Both the rope and tile training sets were similar in their environmental parameters and are, therefore considered together in this section. W here differences do occur, comparisons are made. The principal purpose of analysing the environmental data-sets was to screen the training set sites for unusual or extreme values and to ensure they covered the major gradients under investigation. In addition to phosphorus and

alkalinity it is important to assess the influence of the other parameters which may exert an influence on the observed species assemblages. The distribution of environmental determinants are considered separately below.

Phosphorus

Total phosphorus and filterable reactive phosphorus were highly correlated. From the exploratory analysis it was clear that several of the values obtained for TP were too low (i.e. they were lower than the FRP values). Despite repeated analysis of these samples the source of this error could not be determined. In the majority of cases the error was very low and fell within the experimental errors of the methods. For sample 647 (SLEA l, November sampling), however, the TP value was 29 pgL'^ and FRP was 115 p g L '\ This error was possibly due to adsorption of phosphorus onto the walls of the polyethylene sample bottles prior to analysis, which has been reported in some circumstances (APHS 1989). Alternatively it could have been due to contamination of the FRP sample during analysis. This latter point was thought unlikely because none of the field or laboratory blanks from the same batch of samples showed any trace of contamination. Despite the error this site did not show up as an outlier in the canonical correspondence analysis. It was, however, regarded as a possible problem for further analysis in the development of the transfer function technique.

The sources of phosphorus were almost certainly due to inputs from sewage treatment works and, in more rural areas, leaching from septic tanks. It has been estimated that 93% of the phosphorus exported from the Thames catchment comes directly from domestic STW ’s, with the majority of the remaining 7% coming from diffuse agricultural sources (Roger Sweeting, Environment Agency, pers. comm.). Consequently most of the larger lowland rivers in southern England have high phosphorus concentrations and only the more remote river sites, close to their sources, were found to have TP levels less than 100 p g L '\ The range of phosphorus values in this study is comparable to lowland rivers in other heavily populated regions of Europe (Prygiel et al. 1999).

The principal components analyses identified both TP and FRP as major gradients, highly correlated to the first PGA axes. Phosphorus was therefore considered to be well

represented in the two data-sets. The major difference between the two training sets was that no diatom samples were collected for the tile substratum at the very high TP site of Beverly Brook (site 207). The TP range was, therefore, slightly lower in the tile training set.

Alkalinity

Alkalinity was evenly distributed through the data-sets and was strongly correlated to the second PC A axes. Alkalinity was highly correlated with calcium and was predominantly governed by the underlying geology of the river sites. The highest alkalinity sites were on rivers running though the Hampshire and Wiltshire chalk (e.g. sites 105, 172, 238 & 248) and on the Oolitic and Liassic limestone of the Cotswolds (e.g. sites 9, 12, 13, 14 & 28). The lowest alkalinity sites were on the base-poor Lower Greensands (e.g. sites 162, 177 & 178) and Hasting Beds (e.g. 251, 252, 253 & 262) of Kent and Sussex. The larger river sites and those flowing through the London clays tended to have more average alkalinity (e.g. sites 188, 207, 209 & 214). The values for alkalinity found in this study are typical of agriculturally impacted lowland rivers (Allan 1995)

p H

The majority of sites were circum-neutral to alkaline, which is typical of lowland waters in southern England (Bennion 1993). pH was highly correlated to alkalinity and, like alkalinity, was mainly governed by the underlying geology of the sites. Some of the sites on the very base-poor Lower Greensands may have been expected to have lower pH values, for example Old Lodge stream in the Ashdown Forest (E. Sussex) has a pH range between 4.0-5.0 (Monteith et al. 1997). The sites sampled in this region were, however, all on intensively farmed land and, although representing the lower pH end of the gradient in this study, were between pH 6.7 and 7.1. Agricultural practices have been demonstrated to cause an increase in surface water pH in W ales (Homung et al. 1990) and thus the observed (high) values in this study are less surprising. Despite the relatively low range of values in the data-sets (6.7 - 8.5), the PCA showed pH to be an important variable, correlated with the second axes.

Nitrate

The level of nitrate-nitrogen was relatively high at almost all sites (91 - 10,548 pgL-1,

X = 3,043); this is typical of rivers running though intensively farmed, agricultural areas (Omemik et al. 1980). In the training sets, nitrate concentrations showed no obvious geographical distribution nor did they correlate highly with any other variables in the data-set. This suggests that the major sources of nitrates, at these sites, were due to land use as well as sewage effluent; a situation which is typical of many of lowland England’s freshwaters (Birch & Moss 1990).

Silica

Silica is an important nutrient for diatoms and has been demonstrated to influence the dynamics of diatom populations in lakes (Wetzel 1983). In rivers, however, silica has rarely been found to be in short supply (Lack 1971, Allan 1995), with solubility being higher in circum-neutral to alkaline waters (Marshall 1964). This was found to be the case at all the sample sites, where silica concentrations never dropped below 3.4 m g L '\ From work conducted in lakes, diatom competition with other algal groups has not been found to affect species assemblages until the concentration drops below 0.5 mgL'^ (Wetzel & Likens 1991). Silica was most highly correlated to phosphorus in the training sets, which is likely to be due to STW ’s contributing extra silica, and thus adding to the already high background levels. The European average for silica concentrations in rivers is 6 .8 m g L '\ in this study the mean silica concentration was 10.86 m g L '\ This higher value perhaps reflected the high sewage input into the rivers of southern England.

Conductivity and Ionic Composition

The conductivity and concentration of the major ions (Ca^"^, Na"^, K^, Mg^"^, Cl' & S0 4^’) was relatively high at all sites. The lowest value for conductivity was 130 pS cm'^ and only potassium dropped slightly below 1000 pgL'^ at one site. This is in contrast to many upland waters in the UK, where conductivity and ionic concentrations are generally much lower (Stevenson et al. 1991). The relationship between conductivity and ionic composition was not, however, straightforward. There was a strong correlation between conductivity and calcium (and alkalinity) but calcium was poorly, or negatively, correlated with the other major ions (Tabs. 4.6 & 4.9 and Figs. 4.8 & 4.10), suggesting that a large proportion of these ions (unlike calcium) were not geologically

derived. Potassium, sodium and chloride were most strongly correlated with phosphorus and their source is probably sewage treatment works. The source of magnesium and sulphate was less clear but given significant correlations with both calcium and the other major ions, it is probably of both geological and anthropogenic origin (atmospheric and STW ’s).

Due to the complex ionic structure of the lowland rivers sampled, conductivity could only provide a single measure of ionic strength, rather than information on composition. This was also the reason for the high collinearity between conductivity and the major ions and its eventual removal from the data-sets as a nuisance variable. In effect, by measuring conductivity as well as the major ions, some measurements were effectively being made twice; hence the collinearity.

Flow

As would be expected, many of the lowland river sites were relatively slow flowing, despite active selection for faster flowing reaches. Only 11% of the sites had current velocities greater than 50 cm see"’ and a further 20% had flow rates below the detection limit of the flow meter used (approx. 4.0 cm sec’’). There was little correlation between flow and any other variables. The effects of flow on the diatom communities are discussed below.

Overview

The results from the principal components analyses, for both training sets, suggested that the site selection procedure adequately covered a wide range of lowland river types. W ith respect to phosphorus and alkalinity the PCA identified these variables as major gradients within the selected sample sites and thus provided a suitable range of river sites from which the collection of two diatom training sets could be made, using artificial substrata. With the possible exception of the high end of the TP gradient in the tile training set, there was even coverage of phosphorus and alkalinity in both training sets. The lack of extremely high TP sites in the tile data was not considered as a problem, because an even range of TP from 4-6882 pgL’’ was achieved. Phosphorus concentrations rarely exceed this upper limit in rivers (Allan 1995).

The variability within the other environmental data was high but there were very few sites with extreme values. One site had an unusually high PCA score on axis 3 (site 266, rope training set sample 720) due to very high conductivity (1520 pS cm'^) and concentrations of the major ions, but very low phosphorus concentrations (total P = 16 pgL'^ & FRP = 6 pgL'^). It was noted as a possible outlier but left in the data-set for further analyses. No other sites were identified as outliers with respect to the measured environment.