2. ANÁLISIS PRACTICADO A ÍTEMS PROVENIENTES DEL ICFES
2.2 EJEMPLOS DE PREGUNTAS NÚCLEO COMÚN
2.2.3 Paráfrasis inadecuada.
W ith the use of multivariate numerical techniques it is possible to ascertain the suitability of the two different artificial substrata for the development of predictive, diatom-based, models for the assessment of trophic status. The methods used in this thesis allow for an understanding of ecological patterns to be determined but they are not, however, infallible. They rely entirely on the data provided and thus any taxonomic errors or incorrect environmental measurements will increase the observed errors of the analysis. Likewise it would be naïve to assume that all the influential environmental variables are included in the data-set. The diatom taxa react to the total environment in which they live, which is bound to involve many subtle changes which were either not measured or are impossible to measure accurately. The choice of environmental variables measured for this study was chosen a priori as the most likely to be having a major controlling influence on the diatom assemblages.
Needless to say there are other factors involved; the importance of which were not accounted for in this thesis. For example, grazing by snails, caddisfly larvae and, to a lesser extent, mayfly larvae has been shown to have a marked effect on algal biomass in artificial streams (Lamberti et al. 1987). The extent to which this may affect algal accumulation in rivers is very difficult to assess, but Lamberti and Resh (1983) reported a 5-20 fold increase in algal biomass (not only diatoms) when the caddisfly larvae,
Helicopsyche was excluded. Perhaps of greater concern in this thesis is the evidence that
invertebrate herbivory may result in changes in the species composition due to selective grazing (Sumner & Mclintire 1982, Hill & Knight 1987, 1988, Steinman et al. 1987). Hill & Knight (1988) demonstrated up to a sevenfold over-representation of some diatom taxa in the gut contents of caddisfly larvae {Neophylax). Similarly, the numbers of adnate diatoms were under-represented in the gut contents of mayfly larvae
(Ameletus). The physical disturbance of the micro-sediment layer on the substratum by invertebrates, may also reduce the occurrence of some motile diatom taxa (Hill & Knight 1987). The extent to which these biotic processes were important in determining the training set diatom assemblages can not be tested here. Regardless of these factors, and other unquantified variability, clear signals can be established from the measurements that were made. Phosphorus and alkalinity were the two main environmental variables that accounted for the majority of observed diatom distributions and as such both training sets should be suitable for the development of a predictive model for trophic status.
One further method can be used to ascertain the suitability of a variable for the purposes of environmental prediction. This is to perform CCA, as above, but using only the variable which is to be modelled and exclude all other variables from the analysis. The result is that only the first axis is constrained to the environment; all remaining axes are effectively the same as in DCA (i.e. unconstrained). If the axis 1 (Xi), to axis 2 (X2) ratio
is high, it can be assumed that the variable (TP) explains a sufficiently high proportion of the total variance to be used for prediction (Pienitz et al. 1995). The results for a TP constrained, axis 1 CCA of the two river diatom training sets are summarised in Table 4.20 and 4.21.
A xes
Eigenvalues:
Species-environment correlation: Cumulative percentage variance
o f species data:
Xi'.Xi ratio:
T ab le 4.20 CCA summary statistics o f the rope training set data, with TP as the only variable
1 2 3 4 0.199 0.388 0.295 0.268 0.794 0.0 0.0 0.0 5.5% 0.51 16.1% 24.2% 31.5%
A xes 1 2 3 4 Eigenvalues: 0.214 0.403 0.297 0.275 Species-environment correlations: 0.824 0.0 0.0 0.0
Cumulative percentage variance
o f species data: 5.1% 14.6% 2 1.6% 28.1%
Xi'.X2 ratio: 0.53
T ab le 4.21 CCA summary statistics o f the tile training set data, with TP as the only variable
W here pH has been the focus of environmental reconstruction the ÀiiXz ratio can be as high as 0.93 (Kingston et al. 1992) or 0.84 (Dixit et at. 1991) compared to 0.51 and 0.53 in the data-sets presented here. The response of diatoms to pH, however, is due to direct physiological processes, associated with the pH modification of other water chemistry parameters, e.g. aluminium solubility (Round 1990). This is contrary to the more complex relationships controlling circum-neutral, nutrient rich lowland river diatom assemblages. In other studies on phosphorus the ratio has been found to be much lower, e.g. 0.40 in a British Columbian lake data-set (Hall & Smol 1992) and 0.50 in a South East England pond data-set (Bennion 1993). Dixit et at. (1991) quoted a ratio of > 0.50 for the variable to be suitable for weighted averaging methods of calibration. The Xi:X2
ratios, for the rope and tile training sets, just fall above this criterion and thus these data sets are both considered suitable for the development of a diatom-based model for the assessment of trophic status in lowland rivers.
4.6 C onclusions an d Sum m ary
There was considerable variability in both the diatom assemblages and environment in both training sets. The major environmental gradients determined by principal components analysis were alkalinity and phosphorus and the sites from both training sets conformed well to the a priori selection of these variables. In the rope data, there was a clear species response to alkalinity, identified from detrended correspondence analysis, and despite other variables having a clear influence (i.e. silica, ionic content and manganese), phosphorus showed highly significant correlations with the patterns in species distribution. The tile training set data also showed alkalinity as a dominant factor in shaping the species assemblages, but the role of phosphorus appeared to be greater than in the rope samples, with a higher correlation on the first DCA axis. The importance of flow was also more significant in the tile data, and was attributed to the greater exposure to laminar flow patterns over the flat tile surface.
These patterns were confirmed with the use of direct multivariate gradient analysis (CCA) and with forward selection a subset of variables was identified that independently explained the floristic patterns. Alkalinity was the strongest gradient in the rope data, followed by total phosphorus. The tile samples showed a similar response to phosphorus but the influence of alkalinity was reduced to slightly below that of TP. The choice of using TP as an explanatory variable rather than filterable reactive phosphorus (FRP) had only a very minor statistical advantage and thus either variable is considered suitable for the purposes of this study. With the inclusion of good cation and anion chemistry, conductivity was identified as having excessive collinearity within CCA. This was due to conductivity being dependant on both calcium and the concentration of the other major ion, which were not correlated at all sites. Conductivity was therefore removed from the analysis in both data-sets.
CCA was also used to screen for outlying samples, with unusual environmental-species relations. Only one sample from the rope training set was prominent as a possible outlier (720), but it did not completely satisfy the predetermined criteria for outliers, and was thus left in for further analysis. There were no outliers from the tile sample.
Despite the complexities of diatom responses to their environment, much of which cannot be measured reliably, both the rope and tile substrata yielded diatom assemblages that showed a clear response to phosphorus in lowland rivers. Their suitability was further evaluated by the use of CCA, constrained only to the TP gradient. Both training sets showed X i’X i ratios of > 0.5 and can therefore be considered as appropriate for weighted averaging methods to model phosphorus in lowland rivers. The advantage of one substratum over the other remains unclear. Both substrata yielded different, but equally valid diatom assemblages suitable for modelling. The tiles did have the disadvantage of lower recovery rates but by the placement of duplicates, or an efficient method of securing them to the river bed, it is postulated that this problem could be overcome.