Testing large-scale hypotheses using surveys: the effects of land use on the habitats, invertebrates and birds of Himalayan rivers

Authors

  • S. Manel,

    1. Laboratoire de Biologie des Populations d'Altitude, UMR CNRS 5553, Université Joseph Fourier, BP53-38041 Grenoble Cedex 9, France;
    Search for more papers by this author
  • S.T. Buckton,

    1. Catchment Research Group, School of Biosciences, Cardiff University, PO Box 915, Cardiff CF1 3TL, UK; and
    Search for more papers by this author
  • S.J. Ormerod

    1. Catchment Research Group, School of Biosciences, Cardiff University, PO Box 915, Cardiff CF1 3TL, UK; and
    2. Université de Pau et des Pays de l'Adour, UFR Sciences et Technologie, Campus Montaury, BP 155 F-64601 Anglet, France
    Search for more papers by this author

Dr S. J. Ormerod, Catchment Research Group, School of Biosciences, Cardiff University, PO Box 915, Cardiff CF1 3TL, UK (fax 44 1222874305; e-mail Ormerod@cardiff.ac.uk).

Abstract

1. Piecemeal changes in land use might have cumulative effects on regional biodiversity. However, this hypothesis is difficult to test experimentally at the scales involved, so alternative approaches are required. Here, we illustrate some of the strengths and weaknesses of surveys for evaluating the effects of land use on rivers and river birds over a large area of the Himalayan mountains.

2. We surveyed 180 streams and their catchments in north-west India and Nepal in 1994–96. We then used analysis of covariance (ancova), multiple linear regression and multiple logistic regression to assess how stream habitat structure, stream chemistry, aquatic invertebrate abundance and the occurrence of river birds were affected by land use after accounting for altitudinal pattern.

3. Streams draining terraced catchments differed significantly in habitat structure from other streams. They had more physical modifications, wider channels, fewer cascades, finer substrata and simpler riparian vegetation with fewer trees. We detected no clear effects of land use on stream chemistry, but terracing was accompanied by significantly increased abundances of benthic dipterans, ephemeropterans and total aquatic invertebrates.

4. River bird occurrence was best explained by altitude, and secondarily by habitat structure. Some of the habitat features influenced by terracing significantly affected birds both positively (grey wagtail Motacilla cinerea) and negatively (little forktail Enicurus scouleri, river chat Chaimarrornis leucocephalus, brown dipper Cinclus pallasii, plumbeous redstart Rhyacornis fuliginosus). However, only in the grey wagtail did the presence of terracing per se affect occurrence unequivocally; effects on other species were either small or confounded by altitude.

5. We cannot refute the hypothesis that catchment land use affects Himalayan river ecology, but our data on the regional consequences for river birds were equivocal. We suggest that large-scale surveys, although providing one of the few pragmatic methods of assessing large anthropogenic effects on ecosystems, will need careful design to factor out potential confounds if they are to be used to test hypotheses robustly. They should also be supported where possible with process studies, intervention studies and model applications to independent data.

Introduction

Although experimentation has become a central paradigm in ecology (Hairston 1989), some ecological processes occur at scales too large to be reproduced by manipulation (Levin 1992). These include not only global-scale climate systems or biogeochemical cycles, but also the combined effects of factors generated more locally, such as disturbance (Huston 1999): the component processes might be within the feasible range of replicated ecological experiments (Kareiva & Anderson 1989), but their cumulative regional effects are not. Hypothesis-driven science at large scales therefore presents a major challenge. On the one hand, large-scale surveys offer real insight into regional pattern (Gibbons, Rid & Chapman 1993), and many authors condone the use of such correlative and observational data in hypothesis testing (Caley & Schluter 1997; Gaston & Blackburn 1999). Others, however, maintain that only manipulative experiments are robust enough for this purpose (McArdle 1996; Havens 1999). Indeed, in ecology, correlative surveys have traditionally been defended as methods for generating hypotheses rather than evaluating them.

While debates of this type are often focused on fundamental ecology (Huston 1999), they also have real applied relevance (Ormerod, Pienkowski & Watkinson 1999; Caldow & Racey 2000;Ormerod & Watkinson 2000). For example, current threats to global biodiversity require understanding of how piecemeal or locally generated human activities, such as agriculture, might scale up to affect regional biota. Without the possibility of large-scale manipulation, only observational and modelling approaches can offer insight about large-scale ecological effects. The use of such methods in applied ecology is widespread (Chamberlain et al. 1999; Ferreras & Macdonald 1999; Sanchez-Zapata & Calvo 1999; Henderson et al. 2000) but their strengths and weakness in testing hypotheses have been evaluated only infrequently.

The need to assess ecological change through observation and surveys is particularly acute in remote areas. Biodiversity in the Himalayan mountains, for example, is globally significant due to pronounced endemism, habitat heterogeneity and biogeographic location (Myers 1988, 1990). Important biota in Himalayan rivers include endangered species of freshwater reptiles, cetaceans and fishes (Shrestha 1990; Smith 1993), rich assemblages of benthic organisms (Ormerod et al. 1994; Brewin, Newman & Ormerod 1995; Ormerod et al. 1997; Rothfritz et al. 1997; Suren & Ormerod 1998), and the earth's richest assemblage of specialist river birds (Buckton 1998). However, the effects of human activity on Himalayan river catchments have long been suspected (Eckholm 1975; Ives & Messerli 1989). Initial ideas were that the widespread replacement of semi-natural lands by agriculture would exacerbate monsoonal effects on hill-slope stability, soil erosion, run-off quantity and fluvial character (Eckholm 1975). More recent perspectives suggest local chemical effects on rivers as a result of agricultural intensification (Jenkins, Sloan & Cosby 1995; Collins & Jenkins 1996). Given the spatial scales and remoteness of the areas involved, it is not surprising that quantitative studies of the effects of these changes on river biodiversity are few, and robust hypothesis testing is scarce (Ormerod et al. 1994, 1997; Jüttner, Rothfritz & Ormerod 1996; Suren & Ormerod 1998).

Advances in river catchment ecology now offer opportunities for examining large-scale problems of this type (Johnson & Gage 1997). River basins are seen increasingly as discrete and independent landscape units in which catchment character has strong effects on river quality (Allan & Johnson 1997). These effects are investigated, in turn, through gradient analysis across whole regions, for which many analytical methods are now available (Johnson & Gage 1997). We have recently applied this approach to modelling the distribution of Himalayan river birds over large areas. We showed that models could predict bird distribution effectively in regions independent from those in which they were calibrated (Manel et al. 1999). But is it possible that such correlative approaches can be used to test specific hypotheses about the large-scale effects of catchment land use on Himalayan rivers?

Following Eckholm (1975) and Ives & Messerli (1989), we hypothesized that the presence of intensive terraced agriculture in river catchments should be accompanied by effects on river ecology, and hence on the distribution of specialist species of river birds. We evaluated this hypothesis using data from 180 independent streams in north-west India and Nepal. Using linear regression and analysis of covariance, we aimed to assess the degree to which 46 agricultural streams in this set of 180 contrasted with other streams in habitat character, chemistry and invertebrate abundance. We then used logistic regression to assess whether terracing per se, or variables affected by terracing, had any effects on the presence or absence of river birds. Our approach compliments other recent attempts at appraising large-scale ecological problems (Baillie et al. 2000; Collingham et al. 2000; Cowley et al. 2000; Cresser et al. 2000; Gaston et al. 2000).

Study area

The study involved the same seven regions of the Indian (Kumaon) and Nepali Himalaya described by Manel et al. (1999; Fig. 1). The entire area, of over 1000 km in an east–west direction, forms a major part of the Lesser Himalaya, flanked to the south by the Middle Hills and Terai (Peet et al. 1999). Semi-natural vegetation includes forests of sal Shorea robusta Gaertn, Alnus nepalensis D. Don, chir Pinus roxburghii Sarg, blue pines Pinus wallichiana A. B. Jacks, deodar Cedrus deodara (D. Don), Rhododendron spp. and Quercus spp. At higher altitudes, these are replaced by Picea smithiana (Wall.) Boiss and Abies pindrow (Lamb), giving way to extensive Juniperus and Berberis spp. scrub, and eventually alpine scrub, tundra and boulder fields. However, much natural vegetation has now been replaced by rough seasonal pasture or subsistence crops of maize, barley, wheat, millet, sugarcane and potatoes, grown in mostly rain-fed terraces known locally as bhari. Irrigated terracing for rice production, known locally as khet, is common at lower altitudes (< 2000 m). In Nepal alone, such irrigation covers 1·4 million hectares (53% of all agriculture; Donner 1994), representing a major change in river catchment character. Types of terracing were not distinguished in our survey because both often co-occur on different parts of the same hill-slope so that their separate effects could not be determined. Besides major hill-slope restructuring, terrace cultivation involves applications of pesticides, dung, urea and other nitrogenous fertilizers, some river diversions, water abstraction and alterations in riparian vegetation structure. Effects on stream chemistry include increased solute content (Na, Cl, Si, K, Ca, Mg) due to increased weathering, and increased nutrients due to fertilizer additions (Jenkins, Sloan & Cosby 1995; Collins & Jenkins 1996; Renshaw et al. 1997).

Figure 1.

The study regions in north-western India and Nepal: 1, Roop Kund, n = 19 sites; 2, Pindari, n = 24 sites; 3, Simikhot, n = 24 sites; 4, Dunai, n = 20 sites; 5, Manaslu, n = 29 sites; 6, Makalu, n = 32 sites; 7, Kanchenjunga, n = 32 sites.

Of the 180 catchments in this study, 33% had some terraced agriculture, including 26% (n = 46 sites) with extensive cover (i.e. > 33% catchment cover on one or both banks). Other land uses were scrub or rough pasture (occurring at 69% of sites), broadleaf or mixed forest (70%), conifer (16%), alpine (10%), and sal forest (6%). Tourism and trekking is at a much lower volume in the survey regions than around Annapurna, Everest and Langtang, which currently receive 89% of the visits to Nepal by trekkers (Donner 1994).

Descriptions of the aquatic biota of the study areas can be found elsewhere (Ormerod et al. 1994, 1997; Jüttner, Rothfritz & Ormerod 1996; Rothfritz et al. 1997; Suren & Ormerod 1998).

Methods

All field data were collected in winter, at a time when several species of river birds migrate to the lower altitudes where agriculture is concentrated (Donner 1994). The 180 study streams (n = 19–32 per region), all second to fourth order and in independent catchments, were surveyed in November–December 1994 (regions 3, 4 and 6; Fig. 1), September–October 1995 (regions 1 and 2) and October–November 1996 (regions 5 and 7). This regional pattern of visits was randomized as far as logistically possible to avoid spatio-temporal autocorrelation in the resulting data. Thus, far eastern, far western and central regions were just as likely to be sampled at any phase of the study. In each region, streams were sampled opportunistically when encountered and positions were noted either on maps or using global positioning systems. All procedures were carried out simultaneously in a process lasting 1–2 h per site, with the exception that bird surveyors always arrived at the sites in advance of other surveyors to minimize the effects of disturbance. We aimed to include a wide physico-chemical range of sites so that all major influences on bird distribution might be captured. Streams included those fed by surface run-off, springs, glaciers or seepage.

Environmental data and habitat surveys

At each site, we collected water samples for full ionic analysis as described by Collins & Jenkins (1996). Streams at the time of the study were post-monsoonal, and hence at the most stable phase of the annual hydrograph, so their chemistry was well-represented by single samples (Collins & Jenkins 1996). Habitat structure was assessed using a modified form of the UK river habitat survey (RHS) over 200 m as described elsewhere (Buckton & Ormerod 1997; Manel et al. 1999). The survey records over 120 variables describing the stream channel, flow character, banks and catchment. It first involves 10 ‘spot-checks’ of all the features present at 20-m intervals (Appendix 1; for full methods see Raven et al. 1997). For these it was necessary to add land-use categories typical of the Himalaya (e.g. terraced agriculture, forest of sal, alpine tundra with scrub, boulder fields) to those recorded in Europe (e.g. conifer forest, broadleaf or mixed forest, rough grazing). Individual land uses were coded as absent, scarce or extensive (> 33% of at least one bank). Secondly, a ‘sweep-up’ assessment records predominant habitat features over the whole 200-m survey reach, along with stream and bank dimensions. Altitude and slope were determined on-site using altimeters and clinometers, respectively.

To assess potential prey abundance for birds, we took kick-samples of fixed duration, respectively, from riffles (1 × 2 min) and marginal habitats (1 × 1 min) at each site using a 400-μm mesh net. Invertebrates were separated from coarse debris in each sample by elutriation and flotation on-site before preservation in 70% ethanol for subsequent identification. We obtained a combined count from each habitat (riffles/margins) for all major orders (Plecoptera, Ephemeroptera, Trichoptera, Diptera, Coleoptera, Hemiptera, Odonata). Although semi-quantitative, kick samples provide an assessment of invertebrate abundance across all the aquatic habitats in which birds feed. Where marked variations in invertebrate abundance exist, data collected in this way can correlate strongly with river bird distribution (Logie et al. 1996).

Bird presence/absence

River birds were defined as species with territories centred on the river, feeding directly from it or in the riparian corridor (Table 1). Their presence was recorded in the early morning (07.00–11.00) or late afternoon (15.00–18.00) over the same 200-m reaches involved in river habitat surveys. An evaluation of this choice of method, and a validation exercise at 46 sites, can be found in Manel et al. (1999). River birds are particularly strongly associated with the river corridor, and hence they are more reliably detected than species in other habitats. Moreover, in our case the problem of single visits per site is counterbalanced by the spatial intensity of the survey (i.e. 180 independent sites on 46 replicate terraced catchments and 134 others).

Table 1.  Bird species recorded during surveys of 180 streams in north-west India and Nepal, 1994–96
SpeciesNumber (%) of sites
  1. Note: Species nomenclature follows Inskipp, Linsey & Duckworth (1996) with the exception of our use of river chat Chaimorrornis leuocephalus (Vigors) in preference to their alternative of white-capped water redstart.

Large cormorant Phalcorax carbo3 (1·6)
Ibis bill Ibidorhyncha struthersii2 (1·1)
River lapwing Vanellus cinereus1 (0·5)
Common sandpiper Actitis hypoleucos3 (1·6)
Common kingfisher Alcedo atthis5 (2·7)
Crested kingfisher (CK) Ceryle lugubris7 (3·8)
Grey wagtail (GW) Motacilla cinerea22 (12·2)
Pied wagtail M. alba6 (3·3)
Large pied wagtail M. maderaspatensis4 (2·2)
Brown dipper (BD) Cinclus pallasii54 (30·0)
Plumbeous redstart (PR) Rhyacornis fuliginosus65 (36·1)
River chat (RC) Chaimarrornis leucocephalus59 (32·7)
Blue whistling-thrush (BWT) Myiophoneus caeruleus37 (22·5)
Little forktail (LF) Enicurus scouleri44 (24·4)
Black-backed forktail E. immaculatus1 (0·5)
Slaty-backed forktail (SBF) E. schistaceus9 (5·0)
Spotted forktail (SF) E. maculatus9 (5·0)

All our species nomenclature follows Inskipp, Linsey & Duckworth (1996) with the exception of our use of river chat Chaimorrornis leuocephalus (Vigors) in preference to their alternative of white-capped water redstart. This maintains consistency with our earlier publication (Manel et al. 1999).

Data analysis

All data were independent between sites, because all streams were in distinct subcatchments separated by a minimum of 1–2 km. Prior to any further analysis, habitat variables from RHS were reduced to key variates that described major habitat trends between sites. We used principal components analysis (PCA) on the correlation matrix, separating variables between the ‘sweep-up’ and ‘spot-check’ parts of the survey. Although some of the RHS data are categorical, only ordinal features (minimum three classes; n = 43 variables) were included alongside metric variables (n = 78) in these PCA. The use of ordinal data in PCA parallels their use as dummy variables in multiple regression, and is acceptable (Hair et al. 1995; Jongman, ter Braak & van Tongeren 1995). We excluded measures of slope and altitude, so that the resulting principal components (PC) were expressions of habitat character. Chemical data were also reduced by PCA, with determinands that were undetectable at any given site given nominal values equal to half the detection limit.

We assessed the possible effects of terraced agriculture on habitat and chemical PC using analysis of covariance (ancova). PC scores were compared between site groups coded into those where terraces were scarce/absent (group 1) or extensive (group 2). Altitude was treated as a covariate using general linear modelling procedures. For chemistry, regional effects were possible in view of east–west geological variation across the Himalaya; we thus coded the seven regions separately as categories in a crossed analysis with the two land use categories, also including altitude as a covariable. In all cases, significant effects were tested against F. Requirements for homogeneity of variance, normality of error distributions and pattern among residuals were checked using protocols recommended by Fry (1993), although the use of PC as explanatory variables ensured that analytical criteria were invariably met.

Invertebrate data were transformed by logarithms prior to any analysis to normalize distributions. Because abundances often varied curvilinearly with altitude, we accounted for this effect using multiple regression with quadratic terms and then assessed any additional effects by terraced agriculture added as a dummy variable (coded as above). As before, we followed diagnostic protocols recommended by Fry (1993).

For river birds we first wished to assess which variables best explained the presence of each species. Secondly, we wished to assess whether there were additional significant effects on occurrence due to the presence of terraced agriculture per se once major effects on distribution were partialled out (e.g. due to altitude). Following our previous work (Manel et al. 1999), presence and absence of each species was related to altitude, slope, invertebrate abundance, habitat PC and chemical PC using a generalized linear model: multiple logistic regression with a logit link and binomial error distribution (McCullagh & Nelder 1989; Jongman, ter Braak & van Tongeren 1995). This method performed favourably over artificial neural networks and discriminant analysis for modelling species' distribution (Manel et al. 1999), and is widely used in studies at large spatial scales (Rodriguez & Andren 1999; Suarez, Balbontin & Ferrer 2000).

The logit transformation (equation 1) of the probability of presence/absence (p) was modelled as a linear function of 18 possible explanatory variables (xi, i = 1,18; equation 2), including the presence of terraced agriculture modelled as a dummy variable:

image(eqn 1)
image(eqn 2)

in which b0 and bi are the regression constants. Models were fitted using the maximum likelihood method (McCullagh & Nelder 1989) with backwards elimination to select the final predictor variables (Green, Osborne & Sears 1994; Austin & Meyers 1996). The step function, used in the statistical package Splus4, provides a procedure for this purpose using Akaike's information criterion (AIC); this is a penalized version of the likelihood function in which the best model is given by the lowest value (S-PLUS 1997).

Diagnostic procedures in logistic regression on binary data present particular difficulties. For example, binomial denominators are all equal to one, so that residuals around logistic curves provide a pattern irrespective of whether a fitted model is correct (Collett 1991). In our application, significant variables at each step had to reduce significantly the scaled deviance (equivalent to the residual sum of square in linear regression); the change in scaled deviance as each variable is eliminated is distributed approximately like χ2 (McCullagh & Nelder 1989; Collett 1991). We also used a Wald test, equivalent to testing the ratio (estimate/standard error) of regression parameters (bi) against t. Finally, as stated in the introduction, this modelling approach for predicting presence–absence was tested against independent data from the Himalayan study sites (Manel et al. 1999).

Results

Habitat character and the effects of agriculture

Sites ranged in altitude from 350 to 4695 m above sea level, with slopes from 1° to 30°, and channel widths from less than 1 m to 60 m. Wetted-channel widths on average were only 43% of bank widths. This reflected the braided nature of Himalayan river channels in which discharge at the time of sampling occupied only a proportion of the channel typically filled during monsoon floods (Table 2). As expected, terraced catchments (1445 m ± 1030 SD) were at significantly lower altitudes than other catchment types (2436 m ± 679 SD). Seventy-five per cent of extensively terraced sites were below 1830 m.

Table 2.  The range of site attributes recorded during surveys of 180 streams in north-west India and Nepal in 1994–96
VariableMinimumMaximumMedian
Altitude (m)35046952117
Slope (degree)13510·0
Channel width (m)0·4607·0
Water width (m)0·15323·0
Water depth (m)0·022·20·15
Bank full height (m)0·02301·5
NO3 mg l−10·018·80·09
PO4 mg l−10·020·150·02
Cl mg l−10·12·80·3
Na mg l−10·29·01·3
K mg l−10·0417·91·0
Ca mg l−10·0254·08·5
Mg mg l−10·126·00·9
Si mg l−10·511·53·3
pH6·38·77·7
Conductivity (S cm−1)941362·4

A large array of RHS variables contributed significantly to variations in stream habitat structure between sites (Table 3a,b). For example, major variations from the sweep-up data included a trend from large braided streams with laminar or torrential flow to streams with tree-related features and pool-riffle sequences (Sweep PC1; Table 3a). Similarly, from the spot-checks, Spot PC1 described a trend from streams with various types of channel vegetation (notably bryophytes), leaf litter, complex bank vegetation and fine substrates among boulders to streams with more exposed rocks, boulder and cobble substrate, torrential flow and side- or mid-bars. Further examples of such variations between sites are summarized in Table 3. There were strong correlations between the first (r = −0·72, P < 0·001) and second principal components (r = −0·54, P < 0·001) from the sweep-up and spot-check data corroborating general trends in habitat pattern.

Table 3.  Major trends in habitat structure from principal components analysis of RHS data from 180 streams in north-west India and Nepal in 1994–96. Only variables correlating with each PC at P < 0·001 are displayed, and the percentage of variance explained by each PC is shown in parentheses. + and – signs indicate the direction of change for each variable on each PC
(a) ‘Sweep-up’ data
PC1 (12·1%)
PC2 (9·6%)PC3 (7·2%)PC4 (5·4%)
Shade +Gently sloping banks +Composite banks +Boulder substrata +
Fallen trees +Pebble substrata (%) +Sand substrata (%) + 
Woody debris +Gravel substrate (%) +Vegetated side-bars +Bridges –
Overhanging boughs + Out-falls +Pools –
Exposed roots +Pools –Boulder substrata (%) +Gravel substrata (%) –
Bank-side trees +Overhanging boughs –Silt substrata (%) +Vegetated point bars –
Underwater roots +Water depth – Sand substrata (%) –
Pools +Torrential flow –Irrigation/diversion –Reinforced banks –
RifflesExposed bedrock –Vertical and toe banks –Culverts –
Exposed bedrock +Bank-side trees –Cobble substrata (%) –Unvegetated point bars –
Steep banks +Vertical banks –Reinforced banks –Weirs/dams –
Water width –Underwater roots –Reinforced bank-tops –
Boulder fields –Exposed boulders –Riffles –Resectioned banks –
Water depth –Waterfalls –Bank height –Silt substrata (%) –
Torrential flow –Margin width –Torrential flow –Irrigation/diversion –
Laminar flow –Channel width –Pebble substrata (%) –Out-falls –
Water width –Boulder substrata (%) –Unvegetated point bars – 
Margin width –Bank height –  
Unvegetated mid-channel bars –
Unvegetated side-bars –
Channel width –

One of the most marked effects on habitat structure arose from altitude. Thus, Sweep PC2 (r = 0·50) increased with altitude, with positively scoring sites typically in alpine meadows or boulder fields. Spot PC2 (r = −0·25, P < 0·05) and PC3 (r = 0·39, P < 0·05) also varied weakly but significantly with altitude, indicating that higher sites had simple riparian vegetation, riffles rather than cascades, banks of gravel and pebbles rather than boulders or bedrock, and few anthropogenic modifications (Table 4a); this combination of features occurred often in braided channels influenced by glaciers.

Table 4.  Analysis of covariance assessing effects of terraced agriculture on (a) river habitat structure and (b) stream chemistry after accounting for altitude and regional effects using general linear model procedures; habitat variates are defined in Table 3. (*  P < 0·05; **  P < 0·01; ***  P < 0·001; NS not significant at P < 0·05)
(a) Habitat variateAltitude effects (F1,176)Agriculture effects (F1,176)Least squares mean PC score (with SD)
Extensive terrace (n = 46)Other land uses (n = 134)
Sweep PC10·01 NS5·84*−0·72 (0·36)0·40 (0·24)
Sweep PC267·4***6·46*0·63 (0·28)−0·28 (0·19)
Sweep PC32·41 NS0·37 NS0·18 (0·28)−0·04 (0·19)
Sweep PC410·1432·84***−0·99 (0·21)0·50 (0·14)
Sweep PC51·26 NS0·02 NS−0·01 (0·21)0·02 (0·14)
Spot PC10·73 NS0·14 NS0·11 (0·42)−0·07 (0·23)
Spot PC221·6 ***11·41***−0·92 (0·32)0·40 (0·18)
Spot PC328·12**0·36 NS0·20 (0·21)0·05 (0·11)
Spot PC43·350·10·01 NS−0·05 (0·26)−0·02 (0·14)
Spot PC59·67*2·30 0·10·35 (0·25)−0·10 (0·14)

After accounting for altitudinal effects, there were highly significant differences in habitat character between streams from terraced catchments and colleagues (Table 4a,b). Reduced scores on Sweep PC1, Sweep PC4 and Spot PC2, but increased scores on Sweep PC2, together suggested that terraced streams were typified by wider channels with side- or mid-bars; riffles over finer substrata rather than cascades over boulders; and banks with small slopes, fewer tree-related features and simplified vegetation structure. Terraced sites had several types of modification in the channel and banks, including diversions for irrigation.

Chemistry and agriculture

Reflecting the geological complexity of the Himalaya, stream chemistry varied widely for most determinands (Table 2). Two exceptions were that nitrate concentrations were often low, and phosphate was mostly below detection limits. Base cation concentrations were sometimes very low, but nowhere associated with low pH or apparent acidification (Table 2).

The strongest variations indicated by PCA were in base cations (PC1) and other major solutes (e.g. Na, Cl, K, Si, SO42–; PC2). In turn, there were significant altitudinal and regional variations on these PC, reflecting, for example, generally declining base cations from west to east across the Himalaya. Once such effects were taken into account, no effects by terracing on chemistry were detectable (Table 4b).

Invertebrates and agriculture

Invertebrate communities were typical of hill streams, with large contributions by Ephemeroptera, Plecoptera, Trichoptera, Plecoptera and Diptera; their abundances and taxon richness mostly varied weakly but significantly in a curvilinear fashion with altitude (Fig. 2 and Table 5). After accounting for these altitudinal effects, the abundances of Ephemeroptera (derived least-squared means 85·1 ×/÷ 1·12 SD vs. 141·2 ×/÷ 1·20 SD individuals per sample), Diptera (23·4 ×/÷ 1·09 SD vs. 46·7 ×/÷ 1·2 SD) and total invertebrates (501·2 ×/÷ 1·07 SD vs. 691·8 ×/÷ 1·15 SD) were all greater where there was extensive terracing than at other sites.

Figure 2.

Changes in the abundance of aquatic invertebrate orders with altitude across at 180 streams in north-west India and Nepal in 1994–96. The values are numbers (log10) from pooled kick-samples across riffles and marginal habitats (see text). Regression relationships are in Table 5.

Table 5.  Regressions of invertebrate abundances on altitude and terraced agriculture from 180 streams in north-west India and Nepal in 1994–96. The regressions are of the form y = b0 + b1x1 + b2inline image + Agriculture, in which x1 is altitude (× 10−3), x12 is altitude2 and the presence or absence of extensive agriculture is a dummy variable. The standard errors of each coefficient are given, together with indications of significant differences from 0. Except for taxon richness, all abundances were as log10 numbers per sample (*  P < 0·05; **  P < 0·01; ***  P < 0·001) (see Fig. 2)
 b0 (SE)b1 (SE)b2 (SE)Agriculture (SE)r2F
Taxon richness12·3***(1·9)9·67***(1·75)−2·52***(0·37)NS0·2822·74***
Total invertebrates2·27*** (0·13)0·38*** (0·11)−0·07** (0·02)0·03* (0·02)0·084·82**
Plecoptera0·95***(0·12)0·22***(0·05)NSNS0·1313·28***
Ephemeroptera1·73***(0·17)0·40* (0·16)−0·12***(0·03)0·06* (0·02)0·1711·85***
Trichoptera1·42***(0·16)0·60***(0·14)−0·17***(0·03)NS0·2317·10***
Diptera0·58***(0·17)0·54****(0·15)−0·07* (0·03)0·08** (0·02)0·1812·25***

Bird distribution and agriculture

Seventeen species of river birds were recorded, but some were too thinly dispersed for rigorous assessment of correlates with distribution (Table 2). Thus, we could find no significant correlates with the presence of spotted forktails (n = 9 sites), and only altitude correlated with the presence of crested kingfishers (n = 7 sites). In no cases were chemical variables significant predictors of distribution, despite the wide range of stream chemistry encountered.

In the other species, occurrence was related to altitude in all but one case (brown dipper; Table 6). The occurrences of slaty-backed forktail, plumbeous redstart, river chat and crested kingfisher all increased logistically as altitude declined (Fig. 3b,c,d,g). In little forktail, blue whistling-thrush and grey wagtail, altitudinal distributions were unimodal (Fig. 3a,c,f). After accounting for these altitude effects, habitat character had additional significant effects on the occurrence of eight species, most frequently due to Sweep PC2 and Spot PC1 (Table 6). Thus, brown dipper, plumbeous redstart and river chat all occurred most frequently on larger streams with moderate tree cover, boulder substrates, exposed channel rocks, waterfalls, torrential flows and side- or mid-bars. Slaty-backed forktail showed a converse preference for streams with stable earth banks, finer substrates, pool–riffle sequences, and complex vegetation on the banks and riparian zones.

Table 6.  Results from the stepwise procedure to find the best multiple logistic model relating the presence and absence of Himalayan river birds to altitude, habitat structure (as principal components Sweep PC1–5; Spot PC1–5; see Table 3), chemistry, and the abundances of aquatic benthic invertebrates at 180 streams in north-west India and Nepal in 1994–96. All statistically significant cases are shown where standardized logistic regression coefficients (t= estimate/standard error) indicated effects by the explanatory variables. Values greater than 2 indicate positive effects on species' occurrence (P < 0·05); coefficient less than −2 indicate significant negative effects. Species codes are defined in Table 2. Chemistry, Spot PC3 and Sweep PC3 provided no significant correlates with distribution, and there were no correlates with spotted forktails. Asterisked variables are those significantly affected by terracing (see Table 4b)
 LFSBFBDPRRCBWTGWCK
Altitude −3·64 −5·09−2·79 2·28−2·69
(Altitude)2−4·59    −3·98−2·69 
Sweep PC1    −3·05   
Sweep PC2*  −2·31−3·97−3·25   
Sweep PC4*2·10       
Sweep PC5−2·36       
Spot PC1 −2·703·652·76    
Spot PC2*      −3·66 
Spot PC4     2·66  
Spot PC5    2·06   
Total*    2·10   
Invertebrates
Figure 3.

Logistic regressions relating the probability of occurrence of river birds to altitude at 180 streams in north-west India and Nepal in 1994–96. In each case, alternative logistic relationships (solid symbols) are shown for sites with (+) and without (–) terraced agriculture. The open symbols show sites where species occurred (i.e. P = 1) or were absent (i.e. P = 0) The regression parameters are in Table 6 and the species codes are: LF, little forktail; SBF, slaty-backed forktail; PR, plumbeous redstart; RC, river chat; BWT blue-whistling thrush; GW, grey wagtail; CK, crested kingfisher.

The effects of other habitat were confined to individual species (Table 6). For example, blue whistling-thrush was apparently influenced by Spot PC4, indicating a preference for streams with stable cliffs, boulders/cobble substrata and exposed channel rocks. River chats declined along Sweep PC1, indicating a preference for wider deeper streams with laminar or torrential flows, unvegetated side- or mid-channel bars and wider margins; they avoided streams with dense tree-related features. River chats were the only species to occur more often at sites with a greater total abundance of aquatic invertebrates.

Out of the 12 significant effects on birds due to habitat, five involved characteristics that in turn varied with terraced agriculture (Table 4a). Little forktails tended to avoid streams with reinforced banks and anthropogenic features (weirs, dams, culverts, bridges and out-falls; Sweep PC4). Brown dipper, river chat and plumbeous redstart all declined along Sweep PC2, which increased at terraced sites. Conversely, in grey wagtail, a strong negative influence by Spot PC2 indicated a preference for streams with earth banks, gravel or pebble substrata, more riffles, and resectioned banks and channels. However, only grey wagtail was significantly affected by the presence of extensive terraces per se (difference in scaled deviance = 7·5 > χ2 = 3·88, 1 d.f., P < 0·05): after accounting for altitudinal trends, the probability of occurrence of grey wagtail increased significantly at terraced sites (Fig. 3). In no other case were altitudinal trends in probability of occurrence different between terraced sites and colleagues (Fig. 3).

Overall, the apparent effects of environmental factors on the distribution of Himalayan river birds could be summarized as follows: the effects of altitude were paramount, and a major correlate with the occurrence of seven species. Habitat structure provided a major correlate with the occurrence of brown dippers only, but was a secondary correlate with the occurrence of all other species. Finally, invertebrate abundance (river chat) or terraced agriculture (grey wagtail) had effects detectable on only one species each.

Discussion

Increasingly, the available evidence indicates that terraced agriculture in the Himalayan mountains is accompanied by a range of changes in river ecology. Habitat structure in terraced streams varies substantially from other stream types due to modifications to the channel and riparian zone (this study). Benthic bryophyte cover is reduced in ways consistent with reduced channel stability (Suren & Ormerod 1998). Run-off chemistry differs between terraced streams and colleagues, although this effect is only apparent during the monsoon (Collins & Jenkins 1996; cf. this study) and might be masked by geological controls on solute concentrations (Jenkins, Sloan & Cosby 1995). Nevertheless, algal communities in agricultural catchments are consistent with increased nutrient load and other chemical effects (Jüttner, Rothfritz & Ormerod 1996). Finally, invertebrate abundances are increased in terraced streams (Ormerod et al. 1994; this study), and there might also be effects on invertebrate seasonal phenology (Brewin, Buckton & Ormerod 2000). These patterns parallel catchment-scale effects on rivers due to agriculture in other regions of the world (Johnson et al. 1997; Harding et al. 1999). Elsewhere, agricultural change also continues to have a wide range of effects on terrestrial and aquatic bird species (Duncan et al. 1999; Henderson et al. 2000; Siriwadena et al. 2000). Why, therefore, should the apparent effects on land use on birds in this study be so modest, and confined to increased occurrence in one species?

Not all the effects of terracing on streams will be detrimental to river birds. In previous studies, river bird species feeding on aquatic invertebrates had reduced density where prey were scarce or streams were acidified (Logie et al. 1996; Buckton et al. 1998). By contrast, in this study terracing was accompanied by increased benthic invertebrate abundance, at least in winter, and there is no evidence so far in the Himalaya of acidification (Table 2). Invertebrate numbers in terraced streams decline substantially only during monsoon conditions but recover rapidly once flows decline (Brewin, Buckton & Ormerod 2000). During the monsoon, however, altitudinal migrant birds have already returned to high elevations to breed (see below).

The effect of altered riparian habitat structure on riparian prey abundance is less clearly understood, and warrants attention. Brewin & Ormerod (1994) showed that terrestrial invertebrates made a reduced contribution to invertebrate drift in terraced Himalayan rivers in comparison with rivers draining forest. Even along terraced streams, however, one species was flexible enough to avoid detrimental effects: grey wagtails appeared to be influenced positively by terraced agriculture. Although preferring tree-lined rivers when breeding, in winter they forage opportunistically on a wide range of prey and in a wide range of habitats, switching habitat rapidly under changing circumstances (Ormerod & Tyler 1991). In addition to increased aquatic prey in terraced streams, irrigated and flooded terraces in winter offer an increased range of foraging opportunities. Elsewhere in Asia, grey wagtails show some preference for foraging near to agricultural or urban areas (Hirano 1985), and may well be a human commensal species (Cramp 1988). Such species' gains, like losses, can be indicators of ecological change, and the incidence of grey wagtail at terraced sites shows that altered stream ecology has real consequences for birds.

A further reason why catchment agriculture might not have more marked effects on birds is that they were subsumed by other factors. According to the results of logistic regression, the effects of altitude were the most important influence on most species; this might be expected in species that are altitudinal migrants (Inskipp & Inskipp 1991; Table 7). Although further assessment of the effects of land use on the breeding distribution of river birds is required, such effects might be unlikely given that 75% of terraced sites are below 1800 m, and overlap with breeding distribution will be small (Table 7). Interestingly, the one species not affected by altitude was the brown dipper, which, along with slaty-backed forktail and spotted forktail, had a distribution in this study close to that during breeding (Buckton 1998; Table 7). Dippers are tolerant of low temperatures, being adversely affected only when severe freezing affects the rivers in which they feed (Price & Bock 1983). Here, their distribution most strongly reflected the distribution of coarse substrata, torrential flow, waterfalls and tree cover, consistent with patterns in other Cinclus spp. due to a preference for these features as foraging habitat or nest sites (Price & Bock 1983; Buckton & Ormerod 1997; Buckton et al. 1998).

Table 7.  The altitudinal ranges (m a.s.l.) of some Himalayan river birds during breeding and winter (after Inskipp & Inskipp 1991 * or this study † )
SpeciesAltitudinal range
Summer/breeding*Winter*Winter
Crested kingfisher250–1800250–1800360–1630
Grey wagtail2315–41150–1550530–2400
Brown dipper915–4960915–3100360–4695
Little forktail1830–3000900–1830450–3050
Slaty-backed forktail900–1675900–1675360–1460
Plumbeous redstart1525–442075–2560350–3120
River chat1830–5100245–2590350–3540
Blue whistling-thrush1500–48000–2745360–2660

Problems and possibilities in the survey approach

Despite the patterns in our results, there are difficulties in concluding categorically that terraced agriculture had no effects on river birds other than the grey wagtail. First, effects might occur at more localized scales where catchment modification has been more substantial. In and around the Kathmandu Valley, for example, human population densities are larger than anywhere in Nepal, and intensive terracing has altered stream habitats and water quality to a greater extent (Jüttner, Rothfritz & Ormerod 1996; Collins & Jenkins 1996). The numbers of forktails Enicurus spp. and plumbeous redstarts along Kathmandu Valley rivers are reduced in comparison with adjacent forested streams, which Tyler & Ormerod (1993) attributed to habitat modifications. Our wider-scale survey illustrates how such localized perspectives might not be representative of wider regional pattern across the Himalaya.

Secondly, there was some weak and indirect evidence in our data that little forktail, river chat, brown dipper and plumbeous redstart might be affected negatively by terraced agriculture: some of the habitat features affected by terracing were correlated with these species' occurrence. However, the habitat variable responsible for these effects, Sweep PC2, in turn was the strongest habitat correlate with altitude. Under these circumstances, altitudinal effects on habitat structure, bird distribution and land use were all confounded, and might mask true agricultural effects.

These two features of this study encapsulate the major strengths and weaknesses of the survey approach for addressing large-scale questions in applied ecology: on the one hand, features apparently important locally could be seen in a regional context. On the other, confounding relationships among the resulting data meant that the results were equivocal.

These central strengths and weaknesses of large-scale ecology are of direct relevance to questions in the Himalaya. Despite calls for careful examination using field data or experimentation, the postulated effects of Himalayan land management on rivers have never been examined satisfactorily (Eckholm 1975; Ives & Messerli 1989). Experiments at the scales involved are unrealistic. Site-specific catchment studies would improve understanding of the catchment processes that affect Himalayan rivers and their biota (Hornung & Reynolds 1995), but few have been conducted (Collins & Jenkins 1996). Site-specific studies risk limited representativeness, particularly in such a physically heterogeneous region. They might fail also to capture events with long return periods, such as the intense monsoon floods that might be responsible for major change (Reeve 1996). As indicated in the introduction, they cannot indicate cumulative effects across regions. In the end, therefore, only a landscape approach can match the scale of the Himalayan problem (Ricklefs & Schluter 1993; May & Webb 1994). Extensive surveys provide a pragmatic option for this landscape perspective because they capture realistic variations across sites with contrasting land use, altitude, habitat character and disturbance history. In this study, the survey approach was aided by the use of detailed on-site data collection and subsequent modelling, tools used characteristically in successful catchment studies (Johnson & Gage 1997).

Like many approaches in macroecological research (Blackburn & Gaston 1998), however, catchment surveys are characterized by a range of potential problems. They measure uncontrolled variables, and hence rely on correlation to infer cause and effect. They provide little evidence of the processes responsible for pattern. They are constrained logistically in remote environments, so that site selection cannot be randomized easily. They often reveal confounding patterns, which occurred in our case despite stratified data collection across several regions and altitudes. In our case also, data collection was restricted by logistics to one visit per site, which will have had some consequences for our ability to detect reliably bird presence and absence.

Under these circumstance, ecologists addressing large-scale questions are faced with a major dilemma: either to accept that some problems are beyond the scale of possible investigation, or to address them pragmatically using approaches that fall short of the ideal criteria for experimental design. With the latter the only real option, our strongest recommendation is that large-scale surveys should be designed carefully, or blended with other approaches, if they are to be used to test hypotheses robustly. Possibilities include:

With large-scale ecology such a major focus of attention, these recommendations are likely to have wide currency in applied ecology generally.

Acknowledgements

We thank Dick Johnson and Alan Jenkins of the UK Institute of Hydrology who provided the unmissable opportunity for this fieldwork, funded by the Darwin Initiative for the Survival of Species. The Institute of Hydrology also carried out all the chemical analysis. Our thanks are due to P. A. Brewin for assessing the invertebrate numbers, and to Hem Sagar Bharal and Roger Wyatt for help with bird surveys. We thank Dr Nigel Yoccoz for advice on logistic regression. This paper was written during a visiting professorship by S.J. Ormerod at the Université de Pau et des pays de l'Adour, and we thank Professor Claude Mouchés for providing this important opportunity for collaboration. Dr Nigel Yoccoz, Dr Frank D′ Amico, Dr Ingrid Jüttner, Dr Gill Kerby and the referees provided important comments on the manuscript.

Received 12 October 1999; revision received 24 March 2000

Appendix

Appendix 1

Habitat features recorded by the river habitat survey

In total, 148 variables were available, of which only 27 were not relevant to our sites, or were nominal and therefore unsuitable for data analysis (Raven et al. 1997).

INSTRUMENT READINGSAltitude, slope
PREDOMINANT VALLEY FORMe.g. deep V-shaped valley, asymmetrical floodplain, etc.
RIFFLES, POOLS and POINT BARSNumbers counted
SPOT CHECKSBank material, e.g. bedrock, boulder, cobble, etc.
Presence recorded at 10 pointsBank modifications, e.g. reinforced, resectioned
Bank features, e.g. side-bars, eroding or stable cliffs
Channel substrate, e.g. bedrock, boulder, etc.
Flow type, e.g. free fall (waterfalls), chutes, broken standing waves, smooth
Channel modification, e.g. culverted, resectioned, reinforced, dam, ford
Channel feature, e.g. exposed rocks, mid-channel bars, mature islands
Riparian land use within 5m of bank, e.g. urban, broadleaved woodland, rough pasture, sal Forest,
terraced agriculture, etc.
Bank top and bank face vegetation structure based on presence of bryophytes, short herbs, grasses, tall herbs,
scrub, and trees: bare (none), uniform (1 type), simple (2 or 3 types), complex (4 or more types)
Channel vegetation types, e.g. bryophytes, emergent reeds/rushes, floating
SWEEP-UPLand use within 50m of bank top, as spot-check definitions
Recorded as present or extensive
(> 33% of reach) unless otherwise stated
Bank profiles (natural/modified), e.g. vertical, vertical and toe, steep, gentle, composite (mixture)
Extent of tree and associated features: 6-point scale from none to continuous
extent of channel features, e.g. waterfalls, cascades, riffle pools, runs,
exposed boulders, channel bars, side bars, etc.
DIMENSIONS and INFLUENCESChannel dimensions (m): bank height, bank top width, water width, water depth
Artificial features, e.g. numbers of culverts, weirs, bridges, fords, etc.
Evidence of recent management, e.g. dredging, mowing, weed-cutting
Features of special interest, e.g. braided channel, debris dams, water meadow, bog
Brief description

Ancillary