changes in management initiated during study period
During the 2-year study period, BMPs were adopted at seven of the 30 streams (Table 3); four in the informed farmer group, one by an ‘uninformed’ farmer and two by farmers not involved in the information study. Adopted practices ranged from structural developments such as permanent fencing to exclude stock, to ongoing non-structural changes in management, such as reduction of grazing pressure under wet conditions. Four of the farmers adopted BMPs soon after the March 2001 stream health assessment, two made changes 1 or 2 months before September 2002, and one changed management only 1 month before March 2003 (Table 3). Because of this variation, data for BMP-adopted streams in March 2002 includes three streams at which management had not yet been changed.
Table 3. Details and dates of implementation of BMPs that were voluntarily adopted on seven streams (numbered) to improve water quality and/or overall stream health
|Stream||Change/s in management||Period implemented|
|3a||Buffer strip of at least 5 m left clear of fertilizer application Deer removed from waterway paddocks in wet conditions||Pre-September 2001|
|13a||Deer grazed for less time in waterway paddocks under wet conditions||Pre-September 2001|
|30b||Temporary electric fencing used to exclude deer from stream in paddocks above study paddocks (at first done onlywhen deer were grazing winter crops, but increasingly at other times)||Pre-September 2001|
|21c||Upper 150 m of stream from spring source permanently fenced to exclude stock (note that sample site is a further 235 m downstream in grazed area; buffer width 25 m true right, 7 m true left)||3 months prior to September 2001|
|17a||Waterway in study paddock (250 m) permanently fenced (buffer width average 10 m stream to fence; 10 m section at top of study paddock open for stock water but no grazing during study) dam at top of catchment fenced into smaller paddock for greater grazing control.||5 months prior to September 2002d|
|8a||Waterway in study paddock (200 m) permanently fenced into lane (stream-fence distance 13 m true right, 6 m on true left; deer grazed occasionally within fenced area, but less intensively)||4 months prior to September 2002d|
|Small bog (0·4 Ha) upstream fenced off and drained|| |
|New culvert installed|| |
|5b||Water trough installed in study paddock||1 month prior to March 2003d|
comparison of ‘information’ and ‘no information’ streams
Ordination of the eight parameters of stream health across the 25 streams where landowners were involved in the behavioural study showed no consistency in change when grouped by ‘information’ or by ‘no information’ (Fig. 1), and no difference in the distribution of changes in these two groups (Watson's = 0·0746, P > 0·5).
Figure 1. March 2001–03: PCA ordination of 11 streams on properties of farmers in ‘no information’ group (a) and 14 farmers in ‘information’ group (b) with regard to eight parameters of stream health in (c). All plots obtained from the same ordination. Numbers on arrows in (a) and (b) identify individual streams. Arrowheads indicate position in March 2003 and base position in March 2001. Abbreviations: SHMAK habitat (HAB) and invertebrate scores (INV), nitrate/nitrite-nitrogen (NOx-N), ammoniacal nitrogen (NH4-N), dissolved reactive phosphorus (DRP), total suspended solids (TSS), substrate fines (FINES) and proportion of bare stream bank (BARE). Explained variance: axis 1 = 40·5%, axis 2 = 17·0%.
Download figure to PowerPoint
Trends in INV scores showed a general decrease over time in streams on ‘no information’ farms while ‘information’ stream scores were comparatively stable (Fig. 2a), but there was no significant difference between changes in the two groups (repeated-measures anova interaction term F2,46 = 1·31, P = 0·280). QMCI and MCI scores followed similar and statistically non-significant patterns (F2,46 = 0·43, P = 0·654 for QMCI and F2,46 = 0·087, P = 0·426 for MCI; Fig. 2b,c). HABITAT scores increased more between March 2001 and March 2003 among ‘information group’ than ‘no information’ streams (Fig. 2d), but this difference was again not significant (F2,46 = 0·61, P = 0·549), especially when the background variables DISCHARGE and SDSP (stock days in study paddock) were taken into account (combined covariate effect F2,44 = 7·85, P = 0·001; resulting information effect F2,44 = 0·10, P = 0·907; no farm management variables were significant when included as covariates in any other information analyses).
Figure 2. Means (± SE) of stream health indices from 11 streams on properties of farmers not receiving study information () and 14 streams on properties of informed farmers (x), March 2001–March 2003. (a) INV scores from SHMAK assessment; (b) QMCI scores; (c) MCI scores; (d) HABITAT scores from SHMAK assessment.
Download figure to PowerPoint
comparison of bmp-adopted and reference (no bmp-adopted) streams
There was a clear difference in the direction of changes observed over the two-year sampling period in the seven BMP-adopted streams compared to the reference group of 23, when all streams were ordinated with respect to the same eight stream health measures (Fig. 3a–c). Including background management covariates in the ordination did not improve this distinction.
Figure 3. (a–c) March 2001–03: PCA ordination of seven ‘BMP adopted’ streams (a) and 23 reference streams (b) with regard to the eight stream health parameters in (c). All three plots obtained from the same ordination, using similar data as for Fig. 1 but regrouped into ‘BMP adopted’ and reference streams and including all 30 sites. Numbers on each arrow identify individual streams. Arrowheads indicate position in March 2003, base indicates position in March 2001. See Fig. 1 for abbreviations. Explained variance: axis 1 = 42·2%, axis 2 = 15·0%.
Download figure to PowerPoint
Stream health at BMP-adopted sites changed in a generally consistent manner, shifting towards upper and left-hand quadrants in ordination space between March 2001 and 2003 (Fig. 3a). While trends for some reference streams were similar to those shown by BMP-adopted streams, there was no consistent pattern among these streams (Fig. 3b). Overall, the distribution of change in the two groups was significantly different (Watson's = 0·3012, P < 0·005).
The dominant trend in BMP-adopted sites between years was an increase in HABITAT and INV scores, and a smaller decrease in FINES and BARE (compare Fig. 3a,c). Changes in nutrients and TSS did not appear to be related to the overall trend in treatment streams.
Univariate analyses showed that mean INV scores improved significantly in BMP-adopted streams over the 2-year sampling period, while those in reference streams declined (F2,56 = 9·747, P = 0·001; Fig. 4a). No management covariates altered this relationship significantly (minimum covariate P = 0·207). Because BMP-adopted and reference groups showed significantly different INV scores in March 2001 (T1,28 = 2·286; P = 0·009; note that no significant differences were found for any other ‘pretreatment’ comparisons), the analysis was repeated to compare the seven BMP-adopted streams with seven reference streams paired as closely as possible for initial INV scores (Fig. 4b). A significant positive treatment effect remained (F2,24 = 1·13, P = 0·007).
HABITAT scores from treatment streams improved over the study period, but a less marked positive change was also seen in reference streams, so there was no treatment effect (F2,56 = 1·034, P = 0·362; Fig. 4c). Inclusion of the significant covariates DISCHARGE and SDSP (F2,54 = 9·20, P < 0·001 together) did not reveal evidence for a difference between changes in treatment and reference groups (F2,54 = 0·29, P = 0·747).
Changes in mean QMCI values in both groups over time (Fig. 4d) reflected those seen in INV scores (Fig. 4a), with an overall increase from March 2001 to 2003 in the treatment group, while the mean reference stream QMCI score gradually declined (F2,56 = 2·615, P = 0·082). This group effect was weaker than that seen for INV scores. SDSP was again a significant covariate in an ancova explaining variation in QMCI scores over time (F1,55 = 4·65, P = 0·036; for effect of all other covariates, minimum P = 0·616); when this was taken into account the group × time interaction effect on QMCI scores was somewhat reduced, but remained present (F2,55 = 2·48, P = 0·093).
MCI scores initially declined in both BMP-adopted and reference stream groups (Fig. 4e). While scores from BMP-adopted streams did improve between March 2002 and 2003, there was no significant between-group difference in the overall changes recorded (F2,56 = 0·662, P = 0·520). Inclusion of SDSP (stock days in study paddock, F1,55 = 4·60, P = 0·036; minimum P among all other background variables entered in ancova = 0·085) as a predictor did not reveal any effect of BMP adoption on the MCI (F2,55 = 0·34, P = 0·710).
Inclusion of the three streams in which management had not been changed until after March 2002 (Table 3) can be expected to have made these tests for effects more conservative. Note also that other individual parameters that appeared to drive differences within the ordinations were tested unvariately, but no trends were significant (P > 0·5 in all cases).