**2.7. Statistical analysis**

*2.4.3. BMWP scores for macroinvertebrates families*

44 Water Quality

distribution of the macroinvertebrates.

would be 2, which is the family BMWP bioindication value.

**2.5. Definition of BMWP water quality categories**

and macroinvertebrate community) following [30].

**2.6. BMWP statistical validation**

To assign bioindication values to the different macroinvertebrates families, a data matrix of sampling sites *vs* abundance was constructed using the mean abundance values for each aquatic macroinvertebrate family from all sampling and study sites. The mean abundances were standardized to six abundance classes: class 0 (0 organisms), class 1 (1–3 organisms), class 2 (4–10 organisms), class 3 (11–33 organisms), class 4 (34–100 organisms) and class 5 (>100 organisms). This standardization was carried on following [14] to reduce the possible effect of the overvaluation by the local dominance of some groups due to the nonhomogeneous natural

For each family of macroinvertebrates, the abundance class data were pooled within each *Pcq* interval. In cases where a family appeared in more than one study site within the same *Pcq* interval, the abundance classes of such sites were averaged in order to obtain a single abun‐ dance value per family for each *Pcq* interval. The value of class abundance obtained by a *Pcq* interval indicates the number of times you have to replicate the value of the superior limit corresponding to the *Pcq* interval; for example, if a family obtained a value of 1 for its abun‐ dance class within the 0–1 *Pcq* interval, a 3 for the 3–4 *Pcq* interval, a 4 for the 5–6 *Pcq* interval and a 2 for the 6–7 *Pcq* interval, the data for calculating the fifth percentile would be 1, 4, 4, 4, 6, 6, 6, 6, 7 and 7. The bioindication values for each aquatic macroinvertebrate family were calculated by obtaining the fifth percentile of the abundance class distributions along the *Pcq* intervals where that family was present. This value represents the minimum tolerance value of this family in relation to organic pollution; in the case of this example, the fifth percentile

The water quality category ranges for the BMWP values were assigned following [21]. The median value of the data set of the reference sites was calculated. Scores above this median value will correspond to the "Excellent" quality category, while values that fall between the median and the tenth percentile of that distribution are considered to be in the "Good, not sensible affected" quality category. Values below the tenth percentile were subdivided into four equal parts, which correspond to the categories "Regular,""Bad, polluted,""Bad, very polluted," and "Bad, extremely polluted." The names assigned to each of the water quality categories with some modifications were those proposed by Alba‐Tercedor [29]. The selection of reference conditions included physical, chemical and biological criteria (WQI, *Pcq*, land use

The validation process was performed using three approaches. First, a score prediction test, proposed by Armitage et al. [31], with the average per study site of BMWPobserved *vs* BMWPexpected values. The expected values were calculated with a multiple linear regression (best model procedure) with the qualifying variables for each study site. The observed and expected values were plotted and confidence intervals (*α* = 0.05) were calculated using the XLSTAT software WQI and BMWP values are presented as the mean values of each study site for the four monitoring campaigns. Mean values were also calculated for each study season taking into account the values of all the studied sites. Significant differences between sites and seasons were detected with a bivariate analyses of variance (ANOVA), followed by Student‐Neuman‐ Keuls multiple comparison tests (if the data were normally distributed as well as homosce‐ dastic), or Kruskal‐Wallis test for nonparametric data, both *p* < 0.05, using SigmaPlot version 11.0.
