Freshwater Biology (2010) 55, 2181–2199 doi:10.1111/j.1365-2427.2010.02463.x APPLIED ISSUES Exploring the response of functional indicators of stream health to land-use gradients JOANNE E. CLAPCOTT*, ROGER G. YOUNG*, ERIC O. GOODWIN*AND JOHN R. LEATHWICK† *Cawthron Institute, Nelson, New Zealand † National Institute for Water and Atmospheric Research, Hamilton, New Zealand SUMMARY 1. Broad-scale assessment of stream health is often based on correlative relationships between catchment land-use categories and measurements of stream biota or water chemistry. Few studies have attempted to characterise the response curves that describe how measures of ecosystem function change along gradients of catchment land use, or explored how these responses vary at broad spatial scales. 2. In autumn 2008, we conducted a survey of 84 streams in three bioregions of New Zealand to assess the sensitivity of functional indicators to three land-use gradients: percentage of native vegetation cover, percentage of impervious cover (IC) and predicted nitrogen (N) concentration. We examined these relationships using general linear models and boosted regression trees to explore monotonic, non-monotonic and potential threshold components of the response curves. 3. When viewing the responses to individual land-use gradients, four of five functional indicators were positively correlated with the removal of native vegetation cover and N. In general, weaker and less responsive models were observed for the IC gradient. An analysis of the response to multiple stressors showed d15N of primary consumers and gross primary productivity (GPP) to be the most responsive functional indicators to land-use gradients. The multivariate models identified thresholds for change in the relationship between the functional indicators and all three land-use gradients. Apparent thresholds were <10% IC, between 40 and 80% loss of native vegetation cover and at 0.5 and 3.2 mg L)1 N. 4. The strength of regression models and the nature of the response curves suggest that measures of ecosystem function exhibit predictable relationships with land use. Furthermore, the responses of functional indicators varied little among three bioregions. This information provides a strong argument for the inclusion of functional indicators in a holistic assessment of stream health. Keywords: anthropogenic impacts, cellulose decomposition potential, ecosystem metabolism, wood breakdown, d15N Joanne.Clapcott@cawthron.org.nz ing association between land use and stream condition (Harding et al., 1999; Allan, 2004; Burcher, Valett & Benfield, 2007; Paulsen et al., 2008; Young & Collier, 2009). The advent of geographical information systems and improved resolution of spatial data sets mean it is now possible to predict stream condition based on a desktop analysis of catchment land use (Hawkins et al., 2000; Linke et al., 2007; Moilanen, Leathwick &  2010 Blackwell Publishing Ltd 2181 Introduction Land use and land cover are increasingly used to predict stream health based on the direct and cascadCorrespondence: Joanne E. Clapcott, Cawthron Institute, Private Bag 2, Nelson, New Zealand. E-mail: 2182 J. E. Clapcott et al. Elith, 2008). This approach enables a rapid assessment of the state of the environment that can facilitate broad-scale planning. However, such broad-scale assessments may overlook important characteristics of stream health because of the use of response variables or bioindicators that provide an incomplete assessment of stream properties. In addition, empirical models based on linear relationships between land use and in-stream response variables can result in substantial errors in model predictions when they do not account for interactions among multiple predictors (Allan, 2004; Donohue, McGarrigle & Mills, 2006). Recent advances in empirical modelling techniques, such as boosting, classification and additive regression (Elith, Leathwick & Hastie, 2008; Hastie, Tibshirani & Friedman, 2009) allow for greater understanding of the relationship between land use and indicators of condition. Such techniques can characterise non-linear relationships and identify disturbance thresholds that may lead to a significant change in stream condition (e.g. Donohue et al., 2006). This is advantageous given that several studies have suggested curvilinearity in the functional response of river ecosystems to land-use gradients (Hagen, Webster & Benfield, 2006; Niyogi et al., 2007; Young & Collier, 2009). Measures of ecosystem function, like primary production, respiration and nutrient spiralling (hereafter: ‘functional indicators’), provide an assessment of stream health that complements more traditional approaches based on structural indices that describe biotic communities (Gessner & Chauvet, 2002; Von Schiller et al., 2008; Young, Matthaei & Townsend, 2008). Indeed, studies to date have shown relatively weak correlations between the responses of structural and functional indicators to land use (McKie & Malmqvist, 2009; Young & Collier, 2009). Simultaneous use of both structural and functional indicators may thus be advantageous because this lack of redundancy ensures that a more complete picture of stream health is being captured. Functional indicators may also respond to small changes in land use providing early detection of change in response to disturbance or recovery (Rapport, Costanza & McMichael, 1998). There are several functional measures that have been considered as potential indicators of stream health, such as ecosystem metabolism (Bunn, Davies & Mosisch, 1999; Fellows et al., 2006; Young et al., 2008), rates of organic matter processing (Gessner & Chauvet, 2002; Young et al., 2008), organic matter retention (Quinn, Phillips & Parkyn, 2007) and nutrient cycling (Anderson & Cabana, 2005; Udy et al., 2006; Von Schiller et al., 2008). The use of functional measures as indicators of stream health is based on observed relationships with land use (Sandin & Solimini, 2009). For example, GPP can increase in response to catchment and riparian vegetation clearance (Bunn et al., 1999; Von Schiller et al., 2008). Conversely, ecosystem respiration (ER) can exhibit a curvilinear response to land use with reduced respiration in natural streams having low nutrient concentrations and water temperature (Acuña et al., 2008), as well as reduced respiration in highly impacted streams where sedimentation may block the connection between surface waters and the hyporheic zone (Wilson & Dodds, 2009). Numerous studies have examined the influence of land use on organic matter processing mainly by using estimates of leaf litter breakdown (e.g. Young, Huryn & Townsend, 1994; Molinero, Pozo & Gonzalez, 1996; Benfield et al., 2001; Imberger, Walsh & Grace, 2008). The processing of standardised substrata such as cotton or wood has also been investigated (Tank & Winterbourn, 1995; Boulton & Quinn, 2000; Tiegs et al., 2007; Young & Collier, 2009) as a means to minimise some of the variability observed in leaf litter assays because of differences among leaf species and condition (Niyogi, Simon & Townsend, 2003; Hagen et al., 2006). However, wood and cotton processing, like leaf litter breakdown, is influenced by environmental controls (e.g. temperature and nutrient availability) on microbial processes and can be subject to the same confounding responses to land use as other organic matter processes (Hildrew et al., 1984; Boulton & Quinn, 2000; Tiegs et al., 2007). Stable isotopes of nitrogen (N) reflect both the source and transformation of N (Sebilo et al., 2003; Kendall, Elliott & Wankel, 2007; Diebel & Vander Zanden, 2009) and have been suggested as a surrogate measure of nutrient processing in stream catchments (Udy & Bunn, 2001; Lefebvre et al., 2007). Delta-15N values generally indicate the sources of N entering a stream, for example from precipitation and terrestrial nitrification ()15 to 5&), synthetic fertilisers (c. 0&) or sewerage treatment (5–25&) (Anderson & Cabana, 2006; Kendall et al., 2007). High values (e.g. >15&) have also been correlated to high rates of riparian and in-stream denitrification (Sebilo et al., 2003; Udy et al., 2006). Differences in land use among catchments have been shown to correlate with the d15N signal of water  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2183 Methods (Mayer et al., 2002; Cole et al., 2006; Voss et al., 2006), aquatic plants (Udy & Bunn, 2001; Voss et al., 2006), sediment (Udy et al., 2006; Bunting et al., 2007), invertebrates and fish (Fry & Allan, 2003; Anderson & Cabana, 2005). The d15N of primary consumers reflects assimilated N over a long time period (i.e. weeks to months; Peterson & Fry, 1987) and as such may provide a more time-integrated measure of N sources and processes to assess stream health rather than a snapshot of N concentrations. As suggested earlier, there is rapidly growing literature on the development and potential application of functional measures of stream condition. However, few studies have examined how measures of stream ecosystem function vary across a full gradient of impairment (i.e. a non-categorical assessment of streams of varying land use, but see Niyogi, Lewis & Mc Knight, 2001; Dangles et al., 2004), or in response to multiple stressors. For any indicator to be useful, it must respond unequivocally to an anthropogenic stress (Gessner & Chauvet, 2002), and the form and magnitude of response are important criteria for evaluating the suitability of an indicator (Downes et al., 2002). As such, this study aims to characterise how functional indicators of stream health respond to three land-use gradients: native vegetation cover removal, impervious surface cover and stream N concentration. Site selection and sampling strategy Sampling locations were chosen to provide a distribution of sites across gradients of land-use stress. For each stressor gradient, sites were stratified by stream type as identified in the Freshwater Environments of New Zealand (FWENZ) 100-level classification (Leathwick et al., 2010), hereafter referred to as ‘Group’. The classification is a refinement of the River Environment Classification (Snelder et al., 2004) and provides a biologically optimised hierarchical grouping of stream segments (i.e. the portion of a stream between tributary confluences) structured predominantly by similarities in climate, glacial influence, slope and in-stream habitat characteristics. Sites were further limited to geographical ranges for logistical purposes of sampling only (Fig. 1). Twenty-eight sites were situated in each of the three regions; Bay of Plenty region (FWENZ Group 32), Canterbury plains region (FWENZ Group 2) and South Otago ⁄ Southland region (FWENZ Group 108). Sites were limited by stream order (3rd to 5th, with some 2nd-order streams in the Canterbury region) to minimise the influence of stream size on functional responses (cf. Vannote et al., 1980; Harding et al., 1999). All sampling was initiated in austral summer–autumn (February to April) in 2008 to minimise the influence of seasonal variability Bay of Plenty Canterbury Southland Fig. 1 Map showing location of 84 sampling sites in three regions of New Zealand.  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 2184 J. E. Clapcott et al. on functional response (Uehlinger, 2006; Roberts, Mulholland & Hill, 2007). Functional indicators Ecosystem metabolism. The combination of primary production and ER was estimated using the singlestation, open-channel approach which requires measurement of the natural changes in dissolved oxygen concentration (Owens, 1974; Young & Huryn, 1996). Oxygen concentration and temperature were recorded once every 15 min for at least a 24-h period using a D-Opto logger (Zebra-tech, Nelson, New Zealand) attached to a metal stake and deployed in the thalweg at each site. Light recordings, used to provide an indication of day length for the calculation of metabolism, were taken using an Odyssey light logger (Dataflow, Christchurch, New Zealand) attached to the metal stake and set to record photosynthetically active radiation every 15 min. An estimate of the average depth of each site was calculated using at least five measurements of depth at each of five cross sections spaced at regular intervals upstream of the stake to cover the local variation in channel morphology. Metabolism values were calculated using a spreadsheet model described in Young & Collier (2009). Briefly, mean daily ER and the reaeration coefficient (k) were determined using the night-time regression method (Owens, 1974). The reaeration coefficient and ER rate obtained were then used to determine gross photosynthetic rate over the sampling interval using: GPPt ¼ dO2 =dt þ ER  kD where GPPt is the gross photosynthetic rate (g O2 m)3 s)1) over time interval (t). Coefficients of determination for the night-time regression method resulted in average R2 values of 0.87 (±0.10 SD), giving confidence in the method to estimate reaeration and hence calculate metabolic variables in this study. Daily gross primary production (GPP) was estimated as the integral of all temperature-corrected photosynthetic rates during daylight (Wiley, Osborne & Larimore, 1990). Areal estimates were obtained by multiplying the volume-based estimates by average reach depth (m) which allowed comparison among sites with different depths. Organic matter-processing assays. For wood breakdown (Sticks), birch wood (Betula platyphylla Sukaczev) coffee stirrer sticks (114 · 10 · 2 mm) were labelled, a hole drilled in one end, weighed and tied into groups of five using nylon string with an additional plastic label. Short lengths (1 cm) of drinking straws were used to keep the individual sticks separated. Five groups were deployed at each site for 90 days. Each group of sticks was tethered to a metal stake and deployed in locations to cover the full range in flow types present (i.e. run ⁄ riffle ⁄ pool). Each group of sticks was weighed down to keep them submerged close to the stream bed. Following retrieval, sticks were kept on ice and then frozen until analysis. After thawing, sticks were gently washed and then dried to constant weight in a 60 C forced-draft oven and re-weighed. A set of control sticks was oven-dried to determine the proportional difference between air-dry weight and oven-dry weight, which averaged 90% (range 89–90%). This correction factor was used to estimate the initial oven-dry weights for the sticks that were deployed. Exponential decay coefficients (k) for wood were determined using the equation presented in Petersen & Cummins (1974) using degree days as the time variable. Water temperature was recorded throughout the 90-day deployment period every 15 min using a Hobo pendant logger (Onset, MA, U.S.A.) attached to one of the metal stakes at each site. To measure cellulose decomposition potential (Cotton), unbleached cotton test material (Product no. 222; EMPA, St. Gallen, Switzerland) was cut into strips (35 · 240 mm), and five replicate strips were deployed at each site for seven days. Each strip was attached by nylon thread to a metal stake along with the wooden sticks and weighed down close to the stream bed. Following retrieval, cotton strips were stored flat on ice during transport and then subsequently frozen until analysis. After thawing, the cotton strips were gently washed and dried at 40 C for 24 h in a forced-draft oven. Threads were frayed from each side of the strips leaving a width of 32 mm (100 threads) and then strips were cut into three equal lengths discarding the attachment end. The tensile strength (kg) of each length of strip was measured on a motorised tensometer (Sundoo, Whenzhou, China). The initial tensile strength of the strips (55.8 ± 0.39 kg, SE) was determined using a set of control strips that were wetted in tap water, and then frozen and processed with treatment strips. The loss of tensile  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2185 strength was reported in terms of exponential decay coefficients in the same way as the wooden stick data to allow for comparison between these two measures of organic matter processing. Delta-15N of primary consumers. From each site, primary consumers were hand-picked from benthic samples and processed for d15N values to provide an indication of N processes in the stream catchment (D15N). Primary consumers were mayflies (Deleatidium sp.) assumed to feeding preferentially on biofilms (Parkyn et al., 2005); however, shrimp (Paratya curvirostris Heller) or amphipods (Paracalliope fluviatilis Stebbing) were all that were found at some sites (n = 10) and therefore a combination of all species was collected when present to standardise values to those observed for mayflies. Primary consumers were placed in 90% isopropyl alcohol until analysis. In the laboratory, samples were rinsed with deionised water, and their guts were removed and discarded prior to sample drying at 80 C in a forced-draft oven. Ground samples were analysed in a Dumas elemental analyser interfaced to an isotope mass spectrometer (Europa Scientific Ltd, Crewe, England) to obtain d15N values. The ratio of 15N ⁄ 14N was expressed as the relative per mil (&) difference between the sample and a conventional standard (N2 in air). Environmental variables The percentage of native vegetation cover removal (VegR) in the catchment upstream of each site was described from a satellite imagery-based layer consisting of a selection of data extracted from the New Zealand Land Cover Database 2 (LCDB2 is a public good data set managed by the Ministry for the Environment and Landcare Research, http://koordinates.com/layer/1072-land-cover-database-version-2lcdb2/). The percentage of impervious cover (IC) in the catchment upstream of each site was derived from a spatial layer identifying impervious surfaces such as highways and other roads (both sealed and metalled), railway lines and buildings and provides a measure of urbanisation pressure. Stream N concentration (LogN) was estimated from a leaching model combined with a regionally based regression model, implemented within a catchment framework (Woods et al., 2006). The model, based on empirical data, is under development and therefore requires further validation to  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 assess predictive performance (Cichota & Snow, 2009). The model takes into account soil leaching of N and its transport into rivers and streams via surface and subsurface pathways. The N concentration gradient reflects an intensification in land use; not simply the removal of native vegetation but the replacement of native vegetation with land uses of various intensities, such as production forestry, low-intensity grazing, high-intensity grazing, dairy farming, horticulture and urban development. Values were expressed in milligrams per litre and because of their highly skewed distribution were subject to a log10 transformation. Environmental predictors were extracted from a spatially referenced database used in the development of FWENZ (Leathwick et al., 2010). Additional variables included descriptors of upstream and segment-scale climatic, topographical, geological and morphological characteristics (Table 1). Data analysis The response of functional variables to land-use gradients was examined in three steps. Firstly, general linear models (GLM) were used to identify linear responses between land use (VegR, IC, LogN) and functional indicators (GPP, ER, Sticks, Cotton, D15N). All assumptions were checked using standard graphical methods (Quinn & Keough, 2002), and variables were transformed where necessary to meet the assumptions of normality. Next, boosted regression trees (BRT – Friedman, 2001; Hastie et al., 2009) were used to explore the nature of the response of functional variables to individual land-use gradients. BRT is an advanced regression technique in which the final model is composed of a relatively large number (or ensemble) of terms, each consisting of a simple regression tree, typically containing from one to five rules or splits (the number of branches in a tree). Advantages include the ability to fit complex, non-linear responses, the automatic detection and fitting of interactions between predictors and high levels of predictive performance (Elith et al., 2008). For single land-use BRT models, Group and each of the land-use gradients were fitted singularly as predictors; tree complexity (determining the degree of interaction between variables) was set at two to allow the fitting of interactions between Group and the selected land 2186 J. E. Clapcott et al. Table 1 Description of the land-use gradients and environmental variables including the mean and range of values used in this study Environmental descriptor Vegetation cover removal (VegR) LogN Impervious cover (IC) SegHabitat SegSubstrata SegLowFlow SegFlowStability SegSumT SegTSeas SegShade SegSlope USRainDays USAvgT USSlope USCalcium USHardness USPhosphorus USPeat USLake USGlacier Mean (range) Percentage of native vegetation cover removal in the catchment Stream nitrogen concentration, mg L)1, log-transformed Percentage of impervious cover in the catchment Weighted average of proportional cover of local habitat using categories of 1-still; 2-backwater; 3-pool; 4-run; 5-riffle; 6-rapid; 7-cascade Weighted average of proportional cover of bed sediment using categories of 1-mud; 2-sand; 3-fine gravel; 4-coarse gravel; 5-cobble; 6-boulder; 7-bedrock Mean annual 7-day low flow (m3 s)1), fourth-root transformed Annual low flow ⁄ annual mean flow (ratio) Mean summer air temperature (C) Mean winter air temperature (C), normalised with respect to SegSumT Riparian shade (proportional) Segment slope () square-root transformed Days ⁄ year with rainfall in the catchment >25 mm Average air temperature (C) in the catchment, normalised with respect to SegSumT Average slope in the catchment () Average calcium concentration of rocks in the catchment, 1 = very low to 4 = very high Average hardness of rocks in the catchment, 1 = very low to 5 = very high Average phosphorus concentration of rocks in the catchment, 1 = very low to 5 = very high Area of peat in upstream catchment (proportional) Area of lake in upstream catchment (proportional) Area of glacier in upstream catchment (proportional) use; a random selection of one-fifth of the data was used in cross-validation procedures (i.e. fivefold cross-validation) because of the limited size of data sets (i.e. n < 100), and the learning rate was varied to ensure c. 1000 trees were fitted in the final model. For Sticks, only 145 trees could be integrated before a decline in model performance, regardless of learning rate (suggesting poor predictability of Sticks). If a high proportion of the variability was attributed to Group, then this suggests that metric response may differ among stream types, either in absolute values (Y-intercept) or in the magnitude of response (shape of the fitted function). Therefore, where Group accounted for >10% of the explained deviance, fitted functions for each category were examined to identify any difference in responses. Finally, because land-use stressors very rarely exist in isolation, and in fact often co-vary, the response of functional indicators to all three land-use gradients was examined. For these multivariate land-use BRT models, tree complexity was set at three to allow for higher order interactions between Group and the three 65 (0, 100) )0.05 ()1.40, 1.35) 5 (0, 100) 4.0 (2, 4.6) 3.8 (1.7, 5.7) 0.32 0.18 15.89 0.68 0.42 1.48 5.4 0.21 (0, 18.33) (0.01, 0.52) (12.5, 18.8) ()1.3, 3.0) (0, 0.80) (1, 4.42) (1.6, 25.1) ()4.04, 1.64) 10.8 (0.06, 26.77) 1.72 (0.66, 3.0) 3.07 (1.0, 4.92) 3.0 (1.0, 4.9) 0.001 (0, 0.13) 0.001 (0, 0.17) 0.00001 (0, 0.01) land-use gradients; data were divided into five folds for cross-validation, and the learning rate was tuned to ensure c. 1000 trees were added to the final model. Fitted values from this multivariate model were then used as a fixed offset in a second model to examine the degree to which remaining variation could be explained by additional environmental descriptors (e.g. SegShade, USSlope). All 17 environmental variables were offered into the offset model for each functional indicator. BRT models were assessed using both the percentage of deviance explained and the correlation between raw and fitted values, with both estimates computed using a cross-validation procedure that estimates the performance when predicting to new sites using held out samples (Hastie et al., 2009). In addition, the shapes of the functions fitted by these models were visually inspected, i.e. response curves (Fig. 3) and interaction curves (Fig. 4). BRT analyses were carried out in R version 2.7.2 using the ‘gbm’ library of Ridgeway (2006) supplemented by scripts from Elith et al. (2008).  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2187 Equipment failure and vandalism (i.e. removal of substrates from the stream during incubation) resulted in the reduction in data sets for some functional indicators. Results Single land-use gradient Ecosystem metabolism. Measures of GPP ranged from 0.1–20.2 g O2 m)2 d)1 and ER ranged from 0.7– 37.4 g O2 m)2 d)1 (n = 77). Generalised linear models showed significant positive relationships between VegR and both GPP (R2 = 0.23, P = 0.04) and ER (R2 = 0.39, P < 0.001), between LogN and GPP (R2 = 0.26, P = 0.02) and ER (R2 = 0.30, P = 0.006), but linear relationships with IC were not significant (Fig. 2). In the single land-use BRT models, VegR explained 19.1% of the variance in GPP, and similar responses to LogN and IC were indicated as well (Table 2). Vegetation removal explained 16.7% of the variance in ER, whereas IC explained as little as 0.1% of the total deviance in ER data (Table 2). High cross-validated correlation coefficients (mostly > 0.4) lent confidence to the output of BRT models (Table 2). Organic matter processing. Rates of cellulose decomposition potential (k) ranged from 3 to 135 · 10)4 dd)1 indicative of 2.6 to 76% cotton tensile strength loss over the seven-day incubation period (n = 75). The GLM showed a linear relationship with VegR (R2 = 0.39, P < 0.001) and LogN (R2 = 0.41, P < 0.001, Fig. 2), but no significant relationship with IC. BRT showed that VegR and LogN could explain 32.8 and 25.9%, respectively, of the deviance in Cotton data, whilst no convincing response was seen between Cotton and IC (Table 2). In comparison, the average processing rate of Sticks was twofold less than that for Cotton, with wood breakdown (k) ranging from 0.6 to 4.5 · 10)4 dd)1 indicative of 5.1–40.9% weight loss over the 90-day incubation period (n = 71). Wood breakdown was the poorest of the functional indicators in terms of its relationship with land use. No significant linear relationships were evident (Fig. 2), and BRT models were unable to confidently identify functional curves [i.e. low cross-validation coefficient (CV) values] or explain any of the deviance in Sticks using any of the three land-use gradients (Table 2).  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Delta-15N of primary consumers. The d15N value of primary consumers ranged from 1.03 to 14.73& (n = 64) and showed significant linear relationships with all three land-use gradients: VegR (R2 = 0.58, P < 0.001), LogN (R2 = 0.66, P < 0.001), IC (R2 = 0.33, P = 0.006, Fig. 2). BRT models suggested that D15N was the best performing functional indicator in terms of its relationships with the individual land-use gradients. All three land-use gradients explained a large percentage of the deviance in the d15N data (VegR = 41.2%, LogN = 48.7%, IC = 25.1%), and CVs lent confidence to the output (Table 2). Contribution of group. For most functional indicators, Group contributed <25% of the total deviance explained by BRT. For GPP, however, Group consistently contributed a large proportion of the deviance explained for all models (Table 2). The mean and range in GPP was greater in Southland (FWENZ Group 108) compared to other streams. Multivariate land-use gradients Ecosystem metabolism. The combination of three landuse gradients and Group explained 23.6% of the deviance in the GPP data (Table 3). VegR was the dominant predictor (39.6%), and the response curve showed that GPP increased as VegR increased from 40 to 70% (Fig. 3). GPP also increased in response to LogN values between )0.3 and 0.5 mg L)1 and in response to an increase in IC from 0 to 5% (Fig. 3). The interaction between VegR and LogN showed an increasing influence of LogN when VegR values were low (Fig. 4). Group accounted for 31.8% of the deviance explained in the GPP data suggesting different responses between regions, a result similar to that observed in the single land-use analysis. Using the fitted values as an offset in an additional model of GPP, i.e. also using environmental predictors, increased the deviance explained by 31.3 to 54.9% (Table 3); riparian shade (10.5%) accounted for most of this additional explanation, followed by rock hardness in the upstream catchment (3.1%) and catchment slope (2.7%). Only 13.5% of the deviance in the ER model was explained by three land-use gradients and Group, and Group contributed <2.5%. LogN was the main influencing variable (49.6%), and there was a general increase in ER values between )0.3 and 0.5 mg L)1, 2188 J. E. Clapcott et al. Log (GPP+1) 1.5 1.0 0.5 0.0 1.7 Log (ER+10) 1.6 1.5 1.4 1.3 1.2 1.1 1.0 0.010 Cotton dd–1 0.008 0.006 0.004 0.002 0.000 0.0005 Stick dd–1 0.0004 0.0003 0.0002 0.0001 0.0000 15 D15N 10 5 0 0.0 0.2 0.4 IC 0.6 0.8 1.0 0.0 0.2 0.4 0.6 VegR 0.8 1.0 –0.6 –0.3 0.0 0.3 0.6 LogN Fig. 2 Linear relationships between functional indicators and native vegetation cover removal (VegR), impervious cover (IC) and stream nitrogen concentration (LogN). Functional indicators include gross primary productivity-GPP (g O2 m)2 d)1), ecosystem respiration-ER (g O2 m)2 d)1), cellulose decomposition potential – Cotton (k), wooden stick breakdown – Sticks (k) and the d15N of primary consumers – D15N (&).  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2189 Table 2 Boosted regression tree model output for the response of functional indicators to single land-use gradients including stream category (Group) Functional indicator GPP ER Cotton Sticks d15N N Vegetation cover removal Direction of response % deviance explained by Group Total % deviance explained CV LogN Direction of response % deviance explained by Group Total % deviance explained CV Impervious cover (IC) Direction of response % deviance explained by Group Total % deviance explained CV 77 77 75 71 64 + 11.7 19.1 0.55 + 2.1 16.7 0.41 + [<40] 4.8 32.8 0.56 NR <0.1 <0.1 0.22 + [<40] 4.9 41.6 0.70 + 7.5 14.2 0.43 + 1.1 12.3 0.41 + 5.5 25.9 0.51 NR <0.1 <0.1 0.24 + 4.8 48.7 0.63 + [<10] 10.1 21.3 0.46 + [<10] <0.1 0.1 0.26 + [<10] <0.1 0.1 0.47 NR <0.1 <0.1 0.14 + [<10] 2.5 25.1 0.58 The general direction of the response is indicated as (+) = positive, and gradient restrictions of the relationship are indicated in brackets. For example, gross primary productivity (GPP) increases as IC increases, but only between 0 and 10% IC. CV, cross-validation coefficient; NR, no response was evident. Table 3 Boosted regression tree model output for the response of functional indicators to land use and environmental variables. Model A is the response to three land-use gradients and stream category. Model B uses the values from Model A as a fixed offset and 17 additional environmental predictors Total deviance % Functional indicators n Cross-validation coefficient GPP model A GPP model B Ecosystem respiration (ER) model A ER model B d15N model A d15N model B Cotton model A Cotton model B Sticks model A Sticks model B 77 77 77 0.43 (0.13) 0.87 (0.04) 0.34 (0.08) 23.62 54.94 13.51 77 64 64 75 75 71 71 0.74 0.74 0.82 0.41 0.69 0.08 0.35 51.15 47.55 63.31 19.90 42.62 <1 <1 (0.02) (0.06) (0.03) (0.11) (0.07) (0.13) (0.11) with a distinct non-monotonic relationship (Fig. 3). There was an increase in ER between 0 and 10% IC and a general decrease in ER in response to increasing VegR, with a rapid drop in the functional response above 50% cover (Fig. 3). The interaction of the two most influential variables, LogN and IC, illustrated the dominance of LogN in shaping the distribution of ER data (Fig. 4). An offset model increased ER deviance explained from 13.5 to 51.1% (Table 3), with this attributed mostly to segment slope (8.5%) and ripar 2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Relative deviance % Group VegR LogN IC 31.8 39.6 15.6 13 2.5 21.9 49.6 26 7.3 22 64.3 6.4 9.3 37.2 31.7 21.8 23.8 11.6 40.3 24.3 ian shade (5.6%). CVs of >0.7 indicated a high level of repeatability in these models. Organic matter processing. Neither land-use gradients and Group nor environmental predictors explained a significant proportion of the variation in Sticks data in the multivariate land-use BRT models (Table 3). In contrast, land-use gradients and Group explained 19.9% of the deviance in Cotton data. VegR was the major influencing variable (37.2%), and there was a 2190 J. E. Clapcott et al. 1.0 1.0 0.4 –0.4 0.2 0.4 0.5 (22.9%) 0.6 1.0 (32.9%) 2 32 (2.5%) 5e-04 5e-04 0.2 108 1.0 (49.6%) 5e-04 0.8 32 0.0 –0.5 0.0 (21.9%) 2 –0.3 0.0 0.6 108 (31.8%) –0.3 0.2 –5e-04 0.4 0.5 1.0 0.2 0.4 0.2 0.4 –0.3 0.8 5e-04 0.0 –0.5 0.0 (15.6%) 0.0 0.0 0.4 (26%) Cotton –5e-04 0.6 (39.6%) 0.2 0.4 (13%) 0.0 0.0 0.4 –0.4 0.2 –0.5 0.0 0.5 1.0 –5e-04 0.8 –0.3 Fitted function 0.4 –5e-04 0.0 ER Fitted function 0.0 0.0 –0.4 0.0 –0.4 Fitted function 0.4 0.4 GPP 108 (32.3%) 2 32 (12%) 0.4 (6.4%) IC 0.8 2 0 1 –2 –2 –1 –1 0 0 –1 –2 0.0 1 2 2 1 1 0 –1 –2 Fitted function 2 D15N 0.2 0.6 (22%) VegR 1.0 –0.5 0.0 0.5 1.0 108 2 (64.3%) (7.3%) LogN Group 32 Fig. 3 Fitted functions for the response of functional indicators to the three land-use gradients (IC, VegR, LogN) and stream category (Group) for gross primary productivity (GPP), ecosystem respiration (ER), cellulose decomposition potential (Cotton) and d15N of primary consumers (D15N), with a smoothing span of 0.3. Values in parentheses are the relative percentage of deviance explained by each predictor variable. Rug plots inside the X axes show the distribution of sites, in deciles. rapid increase in cellulose decomposition above 80% removal of native vegetation (Fig. 3). The response to LogN suggested a distinct curvilinear relationship with decreasing cellulose decomposition potential between )0.3 and )0.1 mg L)1 followed by an increase in cellulose decomposition potential up to 0.5 mg L)1. The interaction between these two independent variables showed a greater influence of LogN on the Cotton response when VegR was high (Fig. 4). Cotton also displayed a non-monotonic response to IC with an increase in cellulose decomposition between 0 and 5% IC followed by a gradual decline in values up to 50% (Fig. 3). Group contributed <10% to the model. An offset model increased the Cotton deviance explained by 22.7–42.6% with additional deviance attributed to bed sediment size (3.8%), catchment rain days (3.5%), winter temperature (3.1%) and average catchment temperature (3.0%). Delta-15N of primary consumers. The d15N of primary consumers had the strongest relationship with landuse gradients and Group, with these factors explaining  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2191 ER GPP 20 ER 40 GPP *10 GPP × 10 50 30 15 20 10 10 0 0.0 0.8 1.0 LogN 0.5 Lo g (N 0.0 ) 0.2 –0.5 1.0 0.2 0.6 0.4 IC 0.4 R _ G VE IC VegR 0.5 N) 0.0 og( L 0.6 LogN 0.8 –0.5 0.0 D15N Cotton 10 D15N 9 30 20 8 del15N 10^4 Cot ton * cotton × 104 40 7 6 5 4 10 0.8 1.0 0.6 VE 0.4 G _R VegR 0.5 Lo g 0.5 N) 0.0 og( L LogN 0.2 0.8 1.0 LogN 0.6 (N 0.0 ) 0.2 0.4 R G_ VE VegR –0.5 0.0 0.0 –0.5 Fig. 4 Three-dimensional mesh plots showing the interaction of the two most dominant land-use gradients explaining the deviance in model output for gross primary productivity-GPP (g O2 m)2 d)1), ecosystem respiration-ER (g O2 m)2 d)1), cellulose decomposition potential – Cotton (k) and d15N of primary consumers – D15N (&). 47.5% of the deviance in the d15N data (Table 3). Less than 8% of this total variance was attributable to Group, and the most influential land-use variable was LogN (64.3%). There was a monotonic increase in D15N between )0.5 and 0.5 mg L)1 (Fig. 3). There was also a monotonic increase in D15N in response to increasing VegR with a rapid increase after 70% vegetation removal. However, the interaction plot between these two independent variables further illustrated the dominance of LogN in shaping D15N regardless of VegR (Fig. 4). IC had a very limited influence (6.4%) on the distribution of D15N (Table 3). Addition of environmental predictors in an offset model increased the percentage of variance explained by 15.8%, to 63.3%, with riparian shading (3.4%), winter temperature (2.8%) and catchment rain  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 days (2.1%), the next most influential contributors to the D15N model. Discussion Quantifying the nature of the functional response Our results demonstrate the nature of the response of five different functional indicators of stream health to land-use gradients. The nature of response is characterised by the form of the relationship in terms of direction and shape, and it is also characterised by the strength of association. Whereas previous studies have demonstrated a categorical difference between the function of ‘impacted’ and ‘non-impacted’ streams (e.g. McTammany, Benfield & Webster, 2007; Gücker, 2192 J. E. Clapcott et al. Boëchat & Giana, 2008), this study demonstrates the functional response along a gradient of land-use stressors, individually and combined. Ecosystem metabolism. GPP and ER were positively correlated with the percentage of native vegetation removed from catchments, a result similar to that found by Fellows et al. (2006). Native vegetation removal is associated with riparian degradation and increased light and temperature at the stream bed, which stimulate metabolism (Bott et al., 2006). Other studies have measured metabolism across limited gradients of native vegetation cover and observed some degree of non-linearity because of high variability in metabolism rates around very low and very high levels of native vegetation (Von Schiller et al., 2008; Young & Collier, 2009). For example, in a study of 15 sites, Young & Collier (2009) observed high variability in GPP at five sites with <20% native vegetation. Von Schiller et al. (2008) observed no direct correlation between GPP (or ER) and native vegetation cover in a study of 13 streams, possibly because nine of their streams had more than 90% forest vegetation. In our study, sites were distributed across the native vegetation gradient, and a significant correlation observed. Both GPP and ER had significant responses to N concentration in both single land use and multivariate models as we would expect for any nutrient-limited system in response to increasing nutrient levels. Several categorical studies have identified a correlation between metabolism and stream nutrient status (e.g. Mulholland et al., 2001; Gücker, Brauns & Pusch, 2006) and metabolism and nutrient uptake rates (Hall & Tank, 2003; Meyer, Paul & Taulbee, 2005). In these studies, higher metabolic rates were attributed to a stimulation of organic matter breakdown and primary production because of higher nutrient concentrations and higher uptake rates. In our study, however, GPP had a monotonic response, and ER had a distinct nonmonotonic response to N concentration. The mechanism for the ER response is not evident, but the shape of the ER response may explain why previous studies across categories or limited gradients may have observed both positive and negative relationships between ER and N enrichment in streams. For example, in a study of six streams, GPP and ER were higher in streams with upstream sewage treatment plants or agricultural basins compared to relatively unimpacted counterparts (Hornberger, Kelly & Cosby, 1977). In comparison, in a study of 10 streams, a linear decrease in GPP was observed in response to increasing land-use disturbance intensity (but only in one season of 1 year of a 3-year study), and ER had a negative linear relationship with land-use intensity at all times except for autumn (Houser, Mulholland & Maloney, 2005). We identified a strong positive response in GPP and ER from 0 to 10% IC. It has been consistently shown that increased IC in a catchment leads to a flashier hydrograph, elevated nutrients and temperature and altered channel morphology (see reviews by Paul & Meyer, 2001; Walsh et al., 2005). Yet, recent investigations of the response of stream metabolism to IC (or other indicators of urbanisation) have had very inconsistent results. For example, Bott et al. (2006) observed correlations between GPP and ER and a series of landscape indicators of catchment urbanisation but showed that riparian shade was the most influential variable in a study of 10 streams. After factoring out the influence of shade, they attributed the decrease in GPP and ER to increased contaminant loading. However, two categorical studies identified higher rates of GPP and ER in urban streams compared to forested streams and suggested that effluent and more labile sources of carbon were responsible (Ball et al. 1973 and Paul 1999 in Paul & Meyer, 2001). Furthermore, a study of 23 lowland streams in Japan did not identify a relationship between urban land use and GPP or ER, but instead showed a significant decrease in the GPP ⁄ ER ratio (Iwata et al., 2007). Examination of their data shows an uneven spread of sites across their land-use gradient and high variability in GPP and ER values in streams with catchments <10% urban pressure, which equated to >80% agriculture. This may have contributed to the lack of an observed relationship and is similar to the studies discussed earlier in regard to high metabolic variability at land-use extremes. The potential for metabolism to vary at land-use extremes may be an artefact of viewing the response of metabolism to a single stressor in isolation. At these levels of stress, the single land use of interest may no longer be limiting or influencing the functional response. Instead, additional stressors may shape metabolism (e.g. land clearance versus land-use intensification). This type of relationship was illustrated in the multivariate BRT analyses which showed  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2193 the interaction between a loss of native vegetation cover and N concentration in influencing GPP. At high levels of vegetation removal, sites with higher N had higher GPP (Fig. 4). BRT allows for this examination of the interaction between dependent variables, which is often difficult in small data sets. In summary, the conflicting responses of previous studies of metabolism to land-use gradients appear to be because of investigation across limited gradients confounded by viewing stressors in isolation. The results of this study demonstrate the advantages gained by viewing the response of metabolic variables to a full range in stressor values and in combination. More confidence can be given to non-monotonic responses as being real responses and not simply the result of viewing the functional response across a limited gradient, or the inability to fit a linear model to non-linear data. The interaction between stressors can be investigated using BRT interaction plots (e.g. Fig. 4) to further understand the complexity of metabolic response to multiple land-use gradients. Organic matter processing. Cotton showed a generally monotonic response to the loss of native vegetation, with higher decomposition at higher vegetation removal values. However, Cotton also showed a curvilinear response to N with higher decomposition rates at low and high N concentrations. These results suggest that the variability often observed in organic matter processing could indicate a confounding effect of multiple stressors in addition to a confounding effect of multiple response pathways. Previous studies investigating the response of organic matter processing to a single stressor have shown a categorical difference between the breakdown rates of litter in differing streams. Leaf litter breakdown has been shown to be faster in more agriculturally impacted streams, attributed to elevated nutrient concentrations, higher temperature and a change in the macroinvertebrate community composition (Young et al., 1994; Niyogi et al., 2003). Leaf litter breakdown has also been shown to be faster in more urbanised streams, attributed to higher nutrient status and greater discharge and more abrasive flow (Chadwick et al., 2006; Paul, Meyer & Couch, 2006; Imberger et al., 2008). However, leaf breakdown has been shown to be slower in response to an increased gradient of stream acidification (Niyogi et al., 2001; Dangles et al., 2004), metal pollution (Carlisle &  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Clements, 2005) or through the direct or indirect effects of sedimentation (Sponseller & Benfield, 2001; Niyogi et al., 2003). In terms of our results, it seems likely that land-use stressors are acting antagonistically to influence the pattern observed in cellulose decomposition potential. Others have noted antagonistic responses such as the opposing influences of decreased shredder density versus increased nutrient availability (Huryn et al., 2002; Niyogi et al., 2003) or increased stream acidification versus a change in litter composition (Gessner & Chauvet, 2002). We attempted to minimise some of the variability associated with measures of organic matter processing by using standardised substrata. However, there was still considerable variability observed for both cellulose decomposition potential and wood breakdown in response to land use. It is possible that the deployment of cotton strips (and wooden sticks) across all habitat types present may have contributed to the variability observed, although the standard error within sites was similar to that observed when substrata were deployed in a single habitat type (RG Young, unpublished data). It is also possible that part of this weak response may be because of an average of only 21% loss in cotton tensile strength and 20% mass loss in wooden sticks. A 30% loss is recommended as optimal to detect a treatment effect with 90% confidence, although increased processing rates are associated with increasing variability (Correll et al., 1997). Furthermore, mass loss may be a relatively insensitive measure of organic matter processing because biofilm development may cancel out losses because of decomposition (Tiegs et al., 2007). Our results suggest wood breakdown is an insensitive measure of a change in stream health in response to land use, or land use at the catchment-scale is an insensitive predictor of wood breakdown. Delta-15N of primary consumers. The d15N of primary consumers had a significant monotonic response to all three land-use gradients investigated in this study. Similarly, a recent comprehensive study of 82 streams also identified a significant curvilinear relationship between stream N concentration and the d15N of primary consumers (Anderson & Cabana, 2006). Other studies confirm a large range in d15N values of aquatic primary consumers across differences in catchment land use (Fry & Allan, 2003; Anderson & Cabana, 2005; Moore & Suthers, 2005) and nutrient 2194 J. E. Clapcott et al. enrichment (Bergfur et al., 2009; Diebel & Vander Zanden, 2009). Collectively, these studies suggest that this indicator provides a sensitive and responsive signal for measuring a change in stream condition. Whilst the mechanisms of a response in the d15N of primary consumers were not investigated in this study, an evaluation of recent stable isotope studies provides a conceptual outline. A change in land-use intensity leads to increased nutrient inputs to streams with significantly more enriched isotopes related to livestock manure, sewage and fertiliser practices (Mayer et al., 2002; Anderson & Cabana, 2006; Lefebvre et al., 2007). Increased nutrient concentrations stimulate nutrient uptake rates until saturation (Earl, Valett & Webster, 2006; Simon et al., 2007) and stimulate the development of biofilms which are a preferred food for primary consumers (Parkyn et al., 2005). Such relationships are likely for changes in native vegetation cover and stream N concentration. However, the relationship to IC is less well understood. It is possible that point source inputs and human population pressures confound the response to catchment land use with the addition of depleted sources of N (Anderson & Cabana, 2006). Perhaps, this explains in part why the response to IC was the noisiest relationship for this indicator. Response thresholds In this study, linear analysis allowed us to identify the direction of response of functional indicators to land use, and boosted regression trees highlighted potential thresholds in the form of the response. A recent review identified a series of hypothetical relationships between urbanisation and stream biological condition – linear; stepped upper-to-lower-threshold or linearto-lower-threshold (Walsh et al., 2005). Our results suggest that functional indicators demonstrate all of these responses to individual and combined land-use gradients with distinct threshold values where the response stepped to a change in form. Monotonic increases were observed for GPP and D15N in response to increasing vegetation removal and N concentration. Cotton generally increased in response to increasing vegetation removal but a rapid increase in Cotton values was evident between 80 and 100% compared to a much less rapid increase in Cotton between 40 and 80% vegetation removal. Lower and upper thresholds were evident for some functional indicator responses to N, although only at the very extremes of data distribution. For example, GPP increased monotonically between )0.3 and 0.5 mg L)1, and there were no evident responses outside this gradient, but there were also few sites outside this range. Finally, distinct upper limits were observed for most functional responses to IC at 10%. Previous studies have suggested that specific thresholds of land-use change may exist from which we can predict a change in the response of ecological indicators. For example, Wang, Lyons & Kanehl (2001) illustrated how a change in IC from 8 to 12% in a catchment leads to major changes in stream condition based on fish community metrics. Similarly, Donohue et al. (2006) suggested that the ecological status of rivers based on a biological index was impaired when catchments were >69% agriculture. In contrast, Gergel et al. (2002) suggested 20% urbanisation, and 50% agricultural development were thresholds for change in habitat, biological and water quality indicators. Whilst not measuring functional responses, these studies highlight the potential for critical thresholds in land-use change, which is important for aquatic resource management at the landscape-scale (Groffman et al., 2006; Brenden, Wang & Su, 2008). In comparison, Anderson & Cabana (2006) recently suggested that as little as 5% catchment development was a threshold that indicated N saturation and the delivery of excessive N to waterways affecting N processing. In our BRT analyses, such thresholds were identified through examination of model output. Critical values of land-use change that affect in-stream function were <10% for IC, between 40 and 80% for vegetation removal and in general an unlimited response to N, i.e. the response was constant, when evident, over the range of N concentrations encountered. ‘Noise’ in the functional response When the variability in indicator response is not captured by any of the land-use or environmental predictors, there is greater potential for error in model predictions. For functional indicators, understanding this residual noise is important before predictive models can be confidently applied. For metabolism, for example, factors that could lead to residual noise in the relationship with land use as identified in other studies include flow extremes (Acuña et al., 2004),  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2195 substratum stability (Uehlinger & Naegeli, 1998; Atkinson et al., 2008) and geological variability, determining the potential for significant hyporheic respiration in some streams and not others (Crenshaw, Valett & Webster, 2002; Uzarski et al., 2004). Most, if not all, of these factors are likely to have significant correlations with environmental variables measured in this study. For example, segment low flow and segment flow variability provide measures of flow extremes, and geological variability is captured in upstream estimates of calcium, phosphorus and hardness. However, there are some variables that influence metabolism that are not captured in any of our environmental measures. For example, light reaching the stream bed has been shown to significantly account for variance in metabolism data in unmodified catchments (Mulholland et al., 2001). Currently, our measure of segment shade is based on extrapolation of riparian character from catchment vegetation. Similarly, our measure of stream structure is limited to an index of the dominant habitat and substratum types, and previous studies have demonstrated a strong correlation between metabolism and stream structure (e.g. Gücker et al., 2008). The measurement of environmental variables at multiple scales may also significantly improve the predictability of the functional response because the pathways from catchment-scale land-use pressure to in-stream response can be many and varied (Burcher et al., 2007). Furthermore, reach-specific factors may override the influence of land use. For example, light limitation because of riparian vegetation, valley slope or turbidity can have a greater effect on GPP than land use (Young & Huryn, 1999). Using land use as a predictor for stream health does not take into account stressors or filters at local scales, i.e. near-stream and in-stream factors. Stream processes are a function of several interacting scales of factors, and this study does not attempt to describe the disturbance pathways or mechanisms, although there are clearly multiple intermediate elements that can shape the response of stream health to catchment land use. What this study has shown is that the use of additional environmental variables with land-use variables (i.e. the offset models) on average doubled the percentage of deviance explained in functional indicator response. The additional deviance explained by the inclusion of 17 variables accounts for additional environmental variation not captured by Group.  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Clearly, there is a high potential for environmental variables to contribute to patterns in functional response in addition to the overriding influence of land use. The proportion of deviance explained by Group (as a composite measure of environmental variability) was in general quite low. This suggests that the functional response may vary little by stream type across the range investigated here. Pont et al. (2006) suggested that the most useful measures of stream health are insensitive to natural variability. However, currently, the distribution of functional data is geographically restricted in New Zealand (and elsewhere) compared to a wide distribution of data for more traditional structural variables. This, rather than the non-monotonic nature of some responses, is potentially the most significant factor limiting the inclusion of functional indicators into a holistic assessment of stream health. In conclusion, an increasing amount of empirical research has been conducted in recent years to improve our conceptual understanding of stream functions and how they may be measured to assess stream health. Our results contribute significantly to this work by demonstrating the response of functional indicators across comprehensive gradients of land use using exploratory models. There was a wide range of values for all functional measures, and significant curvilinear relationships were observed for the response of GPP, ER, cellulose decomposition potential and the d15N of primary consumers to land use. Our analyses helped explain some of the conflicting relationships seen in previous studies of stream processes and demonstrated potential thresholds in land use that may lead to changes in stream function. Functional indicators performed well in terms of BRT model output, and this suggests that measures of stream function could be confidently predicted from a combination of land use and environmental variables. Acknowledgments This research was funded by the Foundation of Science and Technology of New Zealand through a Cross Departmental Research Pool project administered by the Department of Conservation (DoC). We are grateful for the support and guidance of technical advisory group members including David Kelly (DoC), Marc Schallenberg (University of Otago), Jon Harding (University of Canterbury), Russell Death 2196 J. E. Clapcott et al. (Massey University), Kevin Collier (University of Waikato) and Vera Power (Ministry for the Environment). We thank DoC staff for support with fieldwork and Cawthron staff who helped process the functional data. We also thank two anonymous reviewers and Richard Johnson for constructive feedback and editorial advice. All experiments were conducted with appropriate approvals and comply with New Zealand law. References Acuña V., Giorgi A., Munõz I., Uehlinger U. & Sabater S. (2004) Flow extremes and benthic organic matter shape the metabolism of a headwater Mediterranean stream. Freshwater Biology, 49, 960–971. Acuña V., Wolf A., Uehlinger U. & Tockner K. (2008) Temperature dependence of stream benthic respiration in an Alpine river network under global warming. Freshwater Biology, 53, 2076–2088. Allan J.D. (2004) Landscapes and riverscapes: the influence of land use on stream ecosystems. Annual Review of Ecology, Evolution and Systematics, 35, 257–284. Anderson C. & Cabana G. (2005) d15N in riverine food webs: effects of N inputs from agricultural watersheds. Canadian Journal of Fisheries and Aquatic Sciences, 62, 333–340. Anderson C. & Cabana G. (2006) Does d15N in river food webs reflect the intensity and origin of N loads from the watershed? Science of the Total Environment, 367, 968–978. Atkinson B.L., Grace M.R., Hart B.T. & Veanderkruk K.E.N. (2008) Sediment instability affects the rate and location of primary production and respiration in a sand-bed stream. Journal of the North American Benthological Society, 27, 581–592. Ball R.C., Kevern N.R. & Haines T.A. (1973) An ecological evaluation of stream eutrophication. Technical Report 36. Institute of Water Research, Michigan State University, East Lansing. Benfield E.F., Webster J.R., Tank J.L. & Hutchens J.J. (2001) Long-term patterns in leaf breakdown in streams in response to watershed logging. International Review of Hydrobiology, 86, 467–474. Bergfur J., Johnson R.K., Sandin L. & Goedkoop W. (2009) Effects of nutrient enrichment on C and N stable isotope ratios of invertebrates, fish and their food resources in boreal streams. Hydrobiologia, 628, 67–79. Bott T.L., Montgomery D.S., Newbold J.D., Arscott D.B., Dow C.L., Aufdenkampe A.K., Jackson J.K. & Kaplan L.A. (2006) Ecosystem metabolism in streams of the Catskill Mountains (Delaware and Hudson water- sheds) and Lower Hudson Valley. Journal of the North American Benthological Society, 25, 1018–1044. Boulton A.J. & Quinn J.M. (2000) A simple and versatile technique for assessing cellulose decomposition potential in floodplain and riverine sediments. Archiv für Hydrobiologie, 150, 133–151. Brenden T.O., Wang L. & Su Z. (2008) Quantitative identification of disturbance thresholds in support of aquatic resource management. Environmental Management, 42, 821–832. Bunn S.E., Davies P.M. & Mosisch T.D. (1999) Ecosystem measures of catchment health and their response to riparian and catchment degradation. Freshwater Biology, 41, 333–345. Bunting L., Leavitt P.R., Gibson C.E., McGee E.J. & Hall V.A. (2007) Degradation of water quality in Lough Neagh, Northern Ireland, by diffuse nitrogen flux from a phosphorus-rich catchment. Limnology and Oceanography, 52, 354–369. Burcher C.L., Valett H.M. & Benfield E.F. (2007) The land-cover cascade: relationships coupling land and water. Ecology, 88, 228–242. Carlisle D.M. & Clements W.H. (2005) Leaf litter breakdown, microbial respiration and shredder production in metal-polluted streams. Freshwater Biology, 50, 380–390. Chadwick M.A., Dobberfuhl D.R., Benke A.C., huryn A.D., Suberkropp K. & Thiele J.E. (2006) Urbanization affects stream ecosystem function by altering hydrology, chemistry, and biotic richness. Ecological Applications, 16, 1796–1807. Cichota R. & Snow V.O. (2009) Estimating nutrient loss to waterways - an overview of models of relevance to New Zealand pastoral farms. New Zealand Journal of Agricultural Research, 52, 239–260. Cole M.L., Kroeger K.D., McClelland J.W. & Valiela I. (2006) Effects of watershed land use on nitrogen concentrations and d15 nitrogen in groundwater. Biogeochemistry, 77, 199–215. Correll R.L., Harch B.D., Kirkby C.A., O’Brian K. & Pankhurst C.E. (1997) Statistical analysis of reduction in tensile strength of cotton strips as a measure of soil microbial activity. Journal of Microbiological Methods, 31, 9–17. Crenshaw C.L., Valett H.M. & Webster J.R. (2002) Effects of augmentation of coarse particulate organic matter on metabolism and nutrient retention in hyporheic sediments. Freshwater Biology, 47, 1820–1831. Dangles O., Gessner M.O., Guerold F. & Chauvet E. (2004) Impacts of stream acidification on litter breakdown: implications for assessing ecosystem function. Journal of Applied Ecology, 41, 365–378. Diebel M.W. & Vander Zanden M.J. (2009) Nitrogen stable isotopes in streams: effects of agricultural  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2197 sources and transformations. Ecological Applications, 19, 1127–1134. Donohue I., McGarrigle M.L. & Mills P. (2006) Linking catchment characteristics and water chemistry with the ecological status of Irish rivers. Water Research, 40, 91– 98. Downes B.J., Barmuta L.A., Fairweather P.G., Faith D.P., Keough M.J., Lake P.S., Mapstone B.D. & Quinn G.P. (2002) Monitoring Ecological Impacts: Concepts and Practice in Flowing Waters. Cambridge University Press, New York. Earl S.R., Valett H.M. & Webster J.R. (2006) Nitrogen saturation in stream ecosystems. Ecology, 87, 3140–3151. Elith J., Leathwick J.R. & Hastie T. (2008) A working guide to boosted regression trees. Journal of Animal Ecology, 77, 802–813. Fellows C.S., Clapcott J.E., Udy J.W., Bunn S.E., Harch B.D., Smith M.J. & Davies P.M. (2006) Benthic metabolism as an indicator of stream ecosystem health. Hydrobiologia, 572, 71–87. Friedman J.H. (2001) Greedy function approximation: a gradient boosting machine. Annals of Statistics, 29, 1189–1232. Fry B. & Allan Y.C. (2003) Stable isotopes in zebra mussels as bioindicators of river-watershed linkages. River Research and Applications, 19, 683–696. Gergel S.E., Turner M.G., Miller J.R., Melack J.M. & Stanley E.H. (2002) Landscape indicators of human impacts to riverine systems. Aquatic Sciences, 64, 118– 128. Gessner M.O. & Chauvet E. (2002) A case for using litter breakdown to assess functional stream integrity. Ecological Applications, 12, 498–510. Groffman P.M., Baron J.S., Blett T. et al. (2006) Ecological thresholds: the key to successful environmental management or an important concept of no practical application? Ecosystems, 9, 1–13. Gücker B., Brauns M. & Pusch M.T. (2006) Effects of wastewater treatment plant discharge on ecosystem structure and function of lowland streams. Journal of the North American Benthological Society, 25, 313–329. Gücker B., Boëchat I.G. & Giana A. (2008) Impacts of agricultural land use on ecosystem structure and whole-stream metabolism of tropical Cerrado streams. Freshwater Biology, 54, 2069–2085. Hagen E.M., Webster J.R. & Benfield E.F. (2006) Are leaf breakdown rates a useful measure of stream integrity along an agricultural land use gradient. Journal of the North American Benthological Society, 25, 330–343. Hall R.O.J. & Tank J.L. (2003) Ecosystem metabolism controls nitrogen uptake rates in streams in Grand Teton National Park, Wyoming. Limnology and Oceanography, 48, 1120–1128.  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Harding J.S., Young R.G., Hayes J.W., Shearer K.A. & Stark J.D. (1999) Changes in agricultural intensity and river health along a river continuum. Freshwater Biology, 42, 345–357. Hastie T., Tibshirani R. & Friedman J. (2009) Elements of Statistical Learning: Data Mining, Inference and Prediction. Springer, New York. Hawkins C.P., Norris R.H., Gerritsen J., Hughes R.M., Jackson S.K., Johnson R.K. & Stevenson R.J. (2000) Evaluation of the use of landscape classifications for the prediction of freshwater biota: synthesis and recommendations. Journal of the North American Benthological Society, 19, 541–556. Hildrew A.G., Townsend C.R., Francis J. & Finch K. (1984) Cellulolytic decomposition in streams of contrasting pH and its relationship with invertebrate community structure. Freshwater Biology, 14, 323–328. Hornberger G.M., Kelly M.G. & Cosby B. (1977) Evaluating eutrophication potential from river community productivity. Water Research, 11, 65–69. Houser J.N., Mulholland P.J. & Maloney K.O. (2005) Catchment disturbance and stream ecosystem metabolism: patterns in ecosystem respiration and gross primary production along a gradient of upland soil and vegetation disturbance. Journal of the North American Benthological Society, 24, 538–552. Huryn A.D., Butz Huryn V.M., Arbuckle C.J. & Tsomides L. (2002) Catchment land-use, macroinvertebrates and detritus processing in headwater streams: taxonomic richness versus function. Freshwater Biology, 47, 401– 415. Imberger S.J., Walsh C.J. & Grace M.R. (2007) More microbial activity, not abrasive flow or shredder abundance, accelerates breakdown of labile leaf litter in urban streams. Journal of the North American Benthological Society, 27, 549–561. Iwata T., Takahashi T., Kazama F. et al. (2007) Metabolic balance of streams draining urban and agricultural watersheds in central Japan. Limnology, 8, 243–250. Kendall C., Elliott E.M. & Wankel S.D. (2007) Tracing anthropogenic inputs of nitrogen to ecosystems. In: Stable Isotopes in Ecology and Environmental Science, 2nd edn. (Eds. R. Michener & K. Lajtha), pp. 375–449. Wiley Blackwell Press, Oxford. Leathwick J.R., Snelder T., Chadderton W.L., Elith J., Julian K. & Ferrier S. (2010) Use of generalised dissimilarity modelling to improve the biological discrimination of river and stream classifications. Freshwater Biology, DOI: 10.1111/j.1365-2427.2010. 02414.x. Lefebvre S., Clement J.-C., Pinay G., Thenail C., Durand P. & Marmonier P. (2007) 15N-nitrate signature in 2198 J. E. Clapcott et al. low-order streams: effects of land cover and agricultural practices. Ecological Applications, 17, 2333–2346. Linke S., Pressey R.L., Bailey R.C. & Norris R.H. (2007) Management options for river conservation planning: condition and conservation re-visited. Freshwater Biology, 52, 918–938. Mayer B., Boyer E.W., Goodale C. et al. (2002) Sources of nitrate in rivers draining sixteen watersheds in northeastern U.S.: Isotopic constraints. Biogeochemistry, 57 ⁄ 58, 171–197. McKie B.G. & Malmqvist B. (2009) Assessing ecosystem functioning in streams affected by forest management: increased leaf decomposition occurs without changes to the composition of benthic assemblages. Freshwater Biology, 54, 2086–2100. McTammany M.E., Benfield E.F. & Webster J.R. (2007) Recovery of stream ecosystem metabolism from historical agriculture. Journal of the North American Benthological Society, 26, 532–545. Meyer J.L., Paul M.J. & Taulbee W.K. (2005) Stream ecosystem function in urbanizing landscapes. Journal of the North American Benthological Society, 24, 602–612. Moilanen A., Leathwick J. & Elith J. (2008) A method for spatial freshwater conservation prioritisation. Freshwater Biology, 53, 577–592. Molinero J., Pozo J. & Gonzalez E. (1996) Litter breakdown in streams of the Aguera catchment: influence of dissolved nutrients and land use. Freshwater Biology, 36, 745–756. Moore S.K. & Suthers I.M. (2005) Can the nitrogen and carbon stable isotopes of the pygmy mussel, Xenostrobus securis, indicate catchment disturbance from estuaries in northern New South Wales, Australia? Estuaries, 28, 714–725. Mulholland P.J., Fellows C.S., Tank J.L. et al. (2001) Interbiome comparison of factors controlling stream metabolism. Freshwater Biology, 46, 1503–1517. Niyogi D.K., Lewis W.M.J. & Mc Knight D.M. (2001) Litter breakdown in mountain streams affected by acid mine drainage: biotic mediation of abiotic controls. Ecological Applications, 11, 506–516. Niyogi D.K., Simon K. & Townsend C.R. (2003) Breakdown of tussock grass in streams along a gradient of agricultural development in New Zealand. Freshwater Biology, 48, 1698–1708. Niyogi D.K., Koren M., Arbuckle C.J. & Townsend C.R. (2007) Stream communities change along a catchment land-use gradient: subsidy-stress responses to a pastoral development. Environmental Management, 39, 213– 225. Owens M. (1974) Measurements of non-isolated natural communities in running waters. In: A Manual on Methods for Measuring Primary Production in Aquatic Environments (Ed. R.A. Vollenweider), pp. 111–119. Blackwell Scientific Publications, Oxford. Parkyn S.M., Quinn J.M., Cox T. & Broekhuizen N. (2005) Pathways of N and C uptake and transfer in stream food webs: an isotope enrichment experiment. Journal of the North American Benthological Society, 24, 955–975. Paul M.J. (1999) Stream ecosystem function along a land use gradient. PhD thesis. University of Georgia, Athens. Paul M.J. & Meyer J.L. (2001) Streams in the urban landscape. Annual Review of Ecology and Systematics, 32, 333–365. Paul M.J., Meyer J.L. & Couch C.A. (2006) Leaf breakdown in streams differing in catchment land-use. Freshwater Biology, 51, 1684–1695. Paulsen S.G., Mayio A., Peck D.V. et al. (2008) Condition of stream ecosystems in the US: an overview of the first national assessment. Journal of the North American Benthological Society, 27, 812–821. Petersen R.C. & Cummins K.W. (1974) Leaf processing in a woodland stream. Freshwater Biology, 4, 343–368. Peterson B.J. & Fry B. (1987) Stable isotopes in ecosystem studies. Annual Review of Ecology and Systematics, 18, 293–320. Pont D., Hugueny B., Beier U., Goffaux D., Melcher A., Noble R., Rogers C., Roset N. & Schmutz S. (2006) Assessing river biotic condition at a continental scale: a European approach using functional metrics and fish assemblages. Journal of Applied Ecology, 43, 70–80. Quinn G.D. & Keough M.J. (2002) Experimental Design and Data Analysis for Biologists. Cambridge University Press, Cambridge. Quinn J.M., Phillips N.R. & Parkyn S.M. (2007) Factors influencing retention of coarse particulate organic matter in streams. Earth Surface Processes and Landforms, 32, 1186–1203. Rapport D.J., Costanza R. & McMichael A.J. (1998) Assessing ecosystem health. Trends in Ecology and Evolution, 13, 397–402. Ridgeway G. (2006) Generalised boosted regression models. Documentation on the R Package ‘gbm’ version 1.5-7, http://www.i-pensieri.com/gregr/gbm.shtml. Roberts B.J., Mulholland P.J. & Hill W.R. (2007) Multiple scales of temporal variability in ecosystem metabolism rates: results from 2 years of continuous monitoring in a forested headwater stream. Ecosystems, 10, 588–606. Sandin L. & Solimini A.G. (2009) Freshwater ecosystem structure-function relationships: from theory to application. Freshwater Biology, 54, 2017–2024. Sebilo M., Billen G., Grably M. & Mariotti A. (2003) Isotopic composition of nitrate-nitrogen as a marker of riparian and benthic denitrification at the scale of the whole Seine River system. Biogeochemistry, 63, 35–51.  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 Stream functional indicators and land-use gradients 2199 Simon K., Niyogi D.K., Frew R.D. & Townsend C.R. (2007) Nitrogen dynamics in grassland streams along a gradient of agricultural development. Limnology and Oceanography, 52, 1246–1257. Snelder T.H., Cattanéo F., Suren A.M. & Biggs B.J.F. (2004) Is the River Environment Classification an improved landscape-scale classification of rivers? Journal of the North American Benthological Society, 23, 580– 598. Sponseller R.A. & Benfield E.F. (2001) Influences of land use on leaf breakdown in southern Appalachian headwater streams: a multiple-scale analysis. Journal of the North American Benthological Society, 20, 44– 59. Tank J.L. & Winterbourn M.J. (1995) Biofilm development and invertebrate colonization of wood in four New Zealand streams of contrasting pH. Freshwater Biology, 34, 305–315. Tiegs S.D., Langhans S.D., Tockner K. & Gessner M.O. (2007) Cotton strips as a leaf surrogate to measure decomposition in river floodplain habitats. Journal of the North American Benthological Society, 26, 70–77. Udy J.W. & Bunn S.E. (2001) Elevated d15N values in aquatic plants from cleared catchments: why? Marine and Freshwater Research, 52, 347–351. Udy J.W., Fellows C.S., Bartkow M.E., Bunn S.E., Clapcott J.E. & Harch B.D. (2006) Measures of nutrient processes as indicators of stream ecosystem health. Hydrobiologia, 572, 89–102. Uehlinger U. (2006) Annual cycle and inter-annual variability of gross primary production and ecosystem respiration in a floodprone river during a 15-year period. Freshwater Biology, 51, 938–950. Uehlinger U. & Naegeli M.W. (1998) Ecosystem metabolism, disturbance, and stability in a prealpine gravel bed river. Journal of the North American Benthological Society, 17, 165–178. Uzarski D.G., Stricker C.A., Burton T.M., King D.K. & Steinman A.D. (2004) The importance of hyporheic sediment respiration in several mid-order Michigan rivers: comparison between methods in estimates of lotic metabolism. Hydrobiologia, 518, 47–57. Vannote R.L., Minshall G.W., Cummins K.W., Sedell J.R. & Cushing C.E. (1980) The river continuum concept. Canadian Journal of Fisheries and Aquatic Sciences, 37, 130–137. Von Schiller D., Marti E., Riera J.L., Robot M., Marks J.C. & Sabater F. (2008) Influence of land use on stream  2010 Blackwell Publishing Ltd, Freshwater Biology, 55, 2181–2199 ecosystem function in a Mediterranean catchment. Freshwater Biology, 53, 2600–2612. Voss m., Deutsch B., Elmgren R., Humborg C., Kuuppo P., Pastuszak M., Rolff C. & Schute U. (2006) Source identification of nitrate by means of isotopic traces in the Baltic Sea catchments. Biogeosciences, 3, 663–676. Walsh C.J., Roy A.H., Feminella J.W., Cottongham P.D., Groffman P.M. & Morgan R.P. II (2005) The urban stream syndrome: current knowledge and the search for a cure. Journal of the North American Benthological Society, 24, 706–723. Wang L., Lyons J. & Kanehl P. (2001) Impacts of urbanisation on stream habitat and fish across multiple spatial scales. Environmental Management, 28, 255–266. Wiley M.J., Osborne L.L. & Larimore R.W. (1990) Longitudinal structure of an agricultural prairie river system and its relationship to current stream ecosystem theory. Canadian Journal of Fisheries and Aquatic Sciences, 47, 373–384. Wilson K.C. & Dodds W.K. (2009) Centimetre-scale stream substrate heterogeneity and metabolic rates. Hydrobiologia, 623, 53–62. Woods R., Elliot S., Shankar U. et al. (2006) The CLUES project: Predicting the effects of land-use on water quality – Stage II. Prepared for Ministry of Agriculture and Forestry. NIWA, Hamilton. p. 109. Young R.G. & Collier K.J. (2009) Contrasting responses to catchment modification among a range of functional and structural indicators of river ecosystem health. Freshwater Biology, 54, 2155–2170. Young R.G. & Huryn A.D. (1996) Interannual variation in discharge controls ecosystem metabolism along a grassland river continuum. Canadian Journal of Fisheries and Aquatic Sciences, 53, 2199–2211. Young R.G. & Huryn A.D. (1999) Effects of land use on stream metabolism and organic matter turnover. Ecological Applications, 9, 1359–1376. Young R.G., Huryn A.D. & Townsend C.R. (1994) Effects of agricultural development on processing of tussock leaf litter in high country New Zealand streams. Freshwater Biology, 32, 413–427. Young R.G., Matthaei C.D. & Townsend C.R. (2008) Organic matter breakdown and ecosystem metabolism: functional indicators for assessing river ecosystem health. Journal of the North American Benthological Society, 27, 605–625. (Manuscript accepted 6 May 2010)