Dispersal of organisms in a patchy stream environment under different settlement scenarios


Nick R. Bond (fax +61 38344 4972; e-mail n.bond@pgrad.unimelb.edu.au).


1. Previous work in streams, where many organisms disperse downstream by drifting in the water column, has suggested dispersal distances could be related to the overall proportion of the stream within dead water zones (DWZ), places such as backwaters, pools or behind large obstacles, where the water is out of the main flow and is stored temporarily. However, dispersal distances might also be influenced by the spatial organization of DWZ within a reach; this would require us to distinguish between reaches containing many small DWZ compared with those having a few large ones in order to understand fully differences in dispersal patterns among reaches differing in this regard.

2. We constructed a spatial model of flow in stream reaches, in which we varied the density and size distribution of DWZ distributed through the reach. We defined flow conditions under which stream organisms, modelled as passive particles, would settle or continue to drift downstream. We then examined the horizontal dispersal profiles of such particles through reaches differing in arrays of DWZ.

3. Our results support those of previous studies, but suggest that the spatial arrangement of DWZ may be just as important as the overall proportion of DWZ within a reach. On a more general note, the importance of patch configuration observed in this study reflects the growing view that a thorough understanding of population and ecosystem processes will require the explicit consideration of spatial pattern in the environment.


Spatial heterogeneity in the environment can strongly influence the distribution and abundance of populations. Spatial patchiness can cause variation in local processes and population dynamics, but can also alter the ability of organisms to disperse across the landscape, from one patch of habitat to another. Numerous studies have illustrated the role that dispersal might play in both local (Amarasekare 1998) and regional population dynamics (Pulliam, Dunning & Liu 1992), and the importance of patchy environments to dispersal processes has made the interaction between these the focus of considerable theoretical and empirical research (Kareiva 1990).

Dispersal in patchy landscapes can be most obviously altered by changing the composition and configuration of the landscape so that organisms must disperse over different distances in order to move between patches (Pulliam et al. 1992). However, there is another way in which patch structure can alter dispersal that has been given far less attention. For propagules that cannot control their own movement, and therefore must disperse passively, such as in the case of most plant and some animal propagules, dispersal patterns are controlled by the speed and direction of the medium they are held in, generally air or water. Consequently, the movement of these propagules is largely dependent on the velocity of the currents (including both speed and direction), with settlement success being largely determined by the availability of suitable settlement sites. In these situations, the distribution of patches as well as topographic features that affect local flow dynamics may alter the dispersal distances and directions of the organisms directly, rather than just the likely outcomes of different dispersal abilities (Lloyd, Daniels & Fish 1990).

Examples such as the dispersal of seeds and of invertebrate larvae moved predominantly by fluids, have been modelled quite successfully as simple exponential decay functions, or diffusion models, with dispersal distance a product of the height of the propagule source and wind speed (Greene & Johnson 1995) or water depth and velocity (McLay 1970; Ciborowski 1983; Reynolds et al. 1990), respectively. Empirical models of the dispersion of pollutants by air (MacDonald, Griffiths & Cheah 1997) and solutes by water (Bencala & Walters 1983; Nepf, Mugnier & Zavistoski 1997), along with studies focusing on the movement of biota (Lancaster & Hildrew 1993), have shown that average transport velocities decrease as the number of obstacles in the landscape increases. As a consequence, settlement rates increase, and the average transport distances of individual particles tend to decrease (Eckman 1990; Lancaster, Hildrew & Gjerlov 1996).

Thus, relatively simple models of dispersal can predict actual dispersal distances reasonably well under controlled conditions, but it is still unclear exactly what factors control drift distances in natural environments (Lancaster et al. 1996). Most studies focus on average transport velocity across the landscape, ignoring local variability. Obstacles, by altering local flow dynamics, provide a link between these two scales but this has received relatively little attention to date (Eckman 1990).

One environment in which to examine this link is streams. Obstacles within stream channels, such as rocks and logs, create dead water zones (DWZ), areas of low or zero current velocity, in which water (and particles) is temporarily stored out of the main flow. As well as influencing average velocity, DWZ are also thought to serve as important settlement sites for some invertebrates, particularly those that cannot swim actively (Elliott 1971; Lancaster et al. 1996). Furthermore, streams are unique in that dispersal is largely biased in one direction by the constant downstream flow of water, which greatly simplifies the task of trying to understand dispersal patterns. A large majority of the organisms found in streams disperse as larvae by drifting downstream (often over a distance of 10s of metres, and possibly further in some cases) in the water column. Different taxa also exhibit a wide range of mobilities, which may affect the degree to which dispersal is reliant upon passive transport. Measurements of drift have shown that travel distance is influenced by taxon (which may relate to swimming ability), size of the individual (which may affect both swimming ability, and also behaviour as a ‘passive’ particle) and current velocity (which influences the distance travelled horizontally before a passive particle reaches the bed; McLay 1970; Elliott 1971; Larkin & McKone 1985).

In a recent paper examining dispersal distances in stream reaches with different hydraulic transport properties Lancaster et al. (1996) predicted that increases in the proportion of obstacles (and therefore DWZ) in the stream should decrease mean water velocity, and thereby decrease mean drift distances. Conversely, they predicted that drift distances should increase in streams where the mean water velocity is high and the proportion of DWZ is low. Finally, they argued that the relationship between organism return rates and current velocity should be similar among streams with similar fractions of DWZ.

While their first two predictions were satisfied, the third was not. In streams with similar fractions of dead water, there were large differences in average drift distances that could not easily be explained. Generalizing results from theoretical and empirical studies across a wide range of ecosystems (Bergelson, Newman & Floresroux 1993; Green 1994; Lancaster & Belyea 1997; Ritchie 1998), we suggest that the configuration of patches in the landscape (in this case DWZ acting as settlement sites) might also have significant effects on dispersal patterns, independent of the total fraction of that patch type in the landscape. Thus, the question we address in this study is whether changes in the spatial configuration of DWZ (their individual size and frequency within a reach), independent of changes in the overall proportion within a reach, also influences average drift distances.

Rather than attempting the difficult task of tackling this question empirically, we developed a simple computer model to simulate dispersal through stream reaches differing in their spatial configuration of DWZ. We addressed two specific questions. (i) Is the average dispersal distance of invertebrates affected by the average proportion of DWZ within different stream reaches? (ii) Is this dependent on the size and therefore also number of DWZ within a reach? We also examined the assumption that DWZ are important as settlement sites, by contrasting the result of allowing settlement to occur anywhere on the streambed, rather than just in DWZ.


The model, written in Borland C++ (Borland 1997), is a stochastic grid-based model. An advantage of grid-based models is that key processes can be represented as a set of rules, rather than as mathematical equations as in traditional reaction-diffusion models. This allows qualitative and poorly understood processes to be incorporated into the model more easily (Jeltsch et al. 1997), and has been used to represent movement directions of stream organisms in previous research (Poff & Nelson-Baker 1997). The overall model consisted of two main components (described separately in detail below). First, a spatial lattice was used to represent stream reaches and to establish the size and proportion of obstacles (which create DWZ) within the landscape. Secondly, individual organisms were introduced into the landscape and allowed to move and settle according to a set of predefined behavioural rules, based on where they were in relation to DWZ and obstacles at each time-step of the model (Fig. 1).

Figure 1.

Flow diagram outlining landscape generation and dispersal behaviour in the model.

Landscape construction

Streams were represented as a two-dimensional lattice, 30 cells wide and 1000 cells long. The spatial grain of the model (i.e. the resolution or cell size), which is defined by the grain at which the biological processes of interest take place, was approximately 25 × 25 cm. The temporal grain of the model was undefined with one model time-step being the time required for an organism to move from one cell to the next. These spatial and temporal grains (cell size and length of time-step) explicitly define the smallest scale at which spatio-temporal heterogeneity is described by the model. Cells can exist in one of three states (or terrain types): flowing water, obstacles or DWZ. Obstacles represent objects such as non-submerged rocks or fallen logs that obstruct the flow of the stream. DWZ are located directly downstream of any obstacle, and their size is directly proportional to that of the obstacle. We carried out a survey of the DWZ in two 25-m long reaches, on the Taggerty and Steavenson rivers, south-eastern Australia, to determine the size, shape and proportion of DWZ likely to be found within natural stream channels. These streams, which are third and fourth order, respectively, are typical of small stony-bedded upland streams in this region. The discharge, average velocity and proportion of DWZ at the Taggerty and Steavenson sites were 0·85 m3 s−1, 0·18 m s−1, 12·1% and 1·25 m3 s−1, 0·32 m s−1, 2·2%, respectively. These estimates fall within the range of those previously published for streams of this size (D’Angelo et al. 1993; Lancaster & Hildrew 1993) and provided values for the parameters included in the model. Within the model two key parameters are used to describe the spatial pattern of any given streamscape: (i) the proportion of the stream occupied by obstacles (and hence DWZ); and (ii) information on the distribution of obstacle sizes (mean and standard deviation) in the stream. Landscapes were randomly generated based on the overall proportion, mean size and standard deviation of the size of obstacles specified in the model (Fig. 2).

Figure 2.

Sample streamscapes at two different proportions of obstacles and with two different obstacle size distributions. (a) 5% obstacles; size (mean ± SD) 2 ± 1 cells; (b) 15% obstacles, size 2 ± 1 cells; (c) 5% obstacles, size 6 ± 1 cells; (d) 15% obstacles, size 6 ± 1 cells.

Organism behaviour

We simulated the movement of organisms using a biased random walk (Turchin 1989). Organisms were assigned simple probabilistic behavioural (movement) rules for the flowing water and DWZ landscape types and were unable to occupy sites already occupied by obstacles. In flowing water, the dominant movement direction was downstream, but there was also a small probability of organisms moving sideways, to represent the turbulent nature of flow in streams. The weightings assigned to these different directions varied depending on the spatial configuration of terrain types in the first-order Moore (i.e. 8-cell) neighbourhood of the organism. Movement was biased towards DWZ, again to represent the way we envisage passive particles to behave in natural stream environments (Fig. 3). At each iteration of the model, and in the absence of any obstacles, there was a probability of 0·7 that an organism would move directly downstream, and a probability of 0·3 that it would move diagonally downstream, 0·15 diagonally left, and 0·15 diagonally right. These basic movement patterns were further modified when the organism's neighbourhood cells contained obstacles and/or DWZ. The possible movement directions, and their associated weightings under different combinations of obstacles and DWZ, are shown in Table 1 and Fig. 3. While these rules oversimplify the complexities of turbulent flow around obstacles in real stream channels, it is neither necessary, nor practical, to attempt to incorporate these complexities into a model in order to address the questions of interest here (Carling 1992).

Figure 3.

Possible directional movements and associated probabilities at all unique combinations of obstacles and DWZ in the Moore neighbourhood of the organisms. Each unique case is as described in Table 1, with possible positions relative to obstacles shown across the page.

Table 1.  A description of the eight truly unique situations recognized in the model that an organism could experience with regard to the terrain type of the eight cells in the Moore neighbourhood. The possible movement directions and weightings for each case are shown in Fig. 3
Case numberDescription*
  • *

    Note that under some circumstances several cases can occur together. To avoid problems with this, case 6 takes priority over case 2, and case 4 overrides case 6.

1Diagonally upstream of a rock
2Diagonally upstream of a DWZ
3Beside a DWZ
4Below a DWZ
5Diagonally below a DWZ
6Upstream of a rock
7Free state
8Trapped among obstacles

Many invertebrate larvae, while they disperse passively, may make active choices about whether or not to disperse again after initial settlement takes place (Fonseca, 1999). The probability of an organism settling once within a DWZ (Ps) was varied in an attempt to simulate this process. Organisms were forced to move sideways when directly upstream of an obstacle, and were forced to settle if they became trapped among several obstacles, and did not move downstream after 20 iterations (time-steps). Although these behavioural rules are a considerable abstraction of the way in which turbulent water and transported particles actually move around obstacles, they allowed us to compare the relative effects of different landscape architectures while still maintaining the benefits of a relatively simple model. We do not interpret these results as a reflection of real dispersal distances and settlement rates, but instead aim to make relative comparisons.

We ran the model with several different scenarios of landscape structure and settlement patterns. The simplest model incorporated landscapes in which there were no obstacles, and hence in which settlement was not confined to DWZ, but instead occurred randomly within the landscape at a fixed probability at each time-step. A range of settlement probabilities was tested. The range of probabilities of random settlement (per iteration) was set such that the cumulative probability of an organism settling by the end of the reach was approximately 0·85, which was the minimum proportion of organisms that settled when settlement occurred only in DWZ. The probabilities of settlement in the random scenario were established in this manner as we could not determine a priori how to make settlement probabilities similar under the conditions of random and restricted (i.e. only in DWZ) settlement. As a next step in model complexity, we ran scenarios in which settlement was again random, but with obstacles and DWZ located in the reach. The most complex scenarios included obstacles and DWZ, and restricted settlement to DWZ only. A complete list of the landscape scenarios that we ran is given in Table 2.

Table 2.  Details of the different scenarios of settlement behaviour and landscape type investigated
LandscapeSettlementPr(settlement)*nProportion DWZ (%)Size DWZ (mean ± SD)
  1. *Note that Pr(settlement) refers to the probability of settlement at each iteration of the model under the scenario of random settlement, but refers to the probability of settlement once in a DWZ in the scenarios where settlement is restricted only to DWZ.

No obstaclesRandom0·007250
Obstacles presentRandom0·00725052 ± 1
0·00825056 ± 1
0·009250152 ± 1
0·010250156 ± 1
Obstacles presentDWZ only0·2525052 ± 1
0·5025056 ± 1
0·75250152 ± 1
1·00250156 ± 1

To explore the effects of both obstacles and DWZ on dispersal patterns we considered four scenarios in which the proportion and size of obstacles (and hence DWZ) were varied. More specifically, we set the proportion of obstacles within a reach to either 5% or 15%, and then, at each of these proportions, we examined situations in which the mean size and standard deviation of DWZ (µ ± 1 SD) were set to 2 ± 1 cells or 6 ± 1 cells, respectively. These values again came from field surveys of DWZ in two natural stream reaches, and are also consistent with previously published estimates (D’Angelo et al. 1993; Lancaster et al. 1996). Because many organisms might move multiple times during their life cycle, we also examined the effects of allowing organisms to leave DWZ once they had been entered. Thus, for each of the landscapes containing DWZ, we compared the results when organisms could move downstream again after initial settlement with probabilities of 0·00, 0·25, 0·50, and 0·75.

The data discussed here are based on replicate landscapes rather than on individual organisms. Each landscape scenario was subject to 250 stochastic replications of uniquely arranged obstacles, and in each replicate the movement of 1000 organisms was tested. Results were synthesized as the grand mean across all scenario replicates and standard deviations are based on the set of scenario replicate means, not the set of organism values from each run of the model. In all cases we concentrate on two statistics to describe dispersal: first, how far downstream, on average, organisms tended to disperse within landscapes of each type, and secondly, the total distance organisms travelled through the landscape (including both lateral and downstream movements). In this model, total distance travelled equates to time spent in the drift (and thus also the number of iterations of the model before settlement occurred). We consider this an important parameter, because drifting invertebrates are likely to be more susceptible to predation by fish (Flecker 1992) or other forms of mortality, a risk that is likely to increase with time spent drifting.


Neutral models

Under the neutral model of random settlement with no obstacles (run essentially to test the model's behaviour), settlement of individual organisms followed a negative exponential distribution, with replicate landscapes distributed according to a Poisson distribution. Although the limited range of settlement probabilities tested here gives an impression of a uniform distribution, as the probability of settling per time-step was increased, average drift distances decreased, as did the variance. Because there were no obstacles to force lateral movement, the downstream distance and total distance travelled were always the same (Fig. 4).

Figure 4.

Mean (± 1 SD) (a) total distance and (b) downstream distance for environments with no obstacles and random settlement. These data are based on landscape replicates, which are means of 1000 replicate organisms through each landscape. Pr(settlement) reflects the probability of an organism settling at random at each iteration of the model.

Inclusion of dwz

When obstacles were included, but settlement occurred randomly across the landscape, both total distance travelled and downstream distance dispersed decreased as the probability of settlement increased from 0·007 to 0·010 (Fig. 5). Differences among landscapes differing in the proportion and size of DWZ were similar at each of the four random settlement probabilities. Increasing the proportion of obstacles from 5% to 15% decreased the mean total distance travelled and also the mean downstream distance dispersed (Fig. 5).

Figure 5.

Mean total and downstream distance (± 1 SD) travelled with obstacles present, but random settlement. Ps equals the probability of settlement at each iteration of the model. Mean obstacle sizes are 2 × 2 cells (white columns) and 6 × 6 cells (hatched columns).

The total distance travelled also increased when obstacle size was increased from two to six, a response that was consistent across all settlement probabilities and at both proportions of obstacles. However, the effects of obstacle size on the distance downstream organisms dispersed were dependent on the overall proportion of obstacles in the landscape. When the proportion of obstacles in the stream was 5%, organisms dispersed further downstream when there were many small DWZ. Conversely, when 15% of the stream contained obstacles, average dispersal distances were greater when obstacles were larger but fewer in number. In all cases, however, changing the proportion of obstacles caused a change in downstream distance travelled of roughly 200 cells, while changing obstacle size caused a change of only 20–25 cells.

Settlement in dwz

Restricting settlement to DWZ had a marked impact on dispersal patterns. Irrespective of probability of settlement, distances travelled (both in total and downstream) were considerably smaller than when settlement was completely random. When we initially ran the model, some scenarios showed extremely high levels of variation between runs. Analysis of the raw model output showed these variances to have been caused, in all cases, by just one or two of the 250 replicate landscapes exhibiting obstacle configurations in which most organisms settled very quickly, but a small number made significant lateral movements as they dispersed downstream. We suspect that these ‘rare’ (< 1% of cases for any landscape type) cases, although of potential biological importance, could possibly have occurred under any of the combinations of obstacle proportion and size, but with a very low probability. Thus, in our analysis of total distance travelled, we exclude replicates in which organisms moved more than 1500 cells (Fig. 6). Upon excluding these cases, the results became quite clear-cut. At a given proportion of DWZ within a reach, increasing the average size of DWZ from two to six approximately doubled the total distance travelled, and for a given mean obstacle size, increasing the proportion of obstacles from 5% to 15% approximately halved the total distance travelled. The important point about this is that, in comparison to the situation considered above in which settlement was random, the relative effect of obstacle (and DWZ) size approached that associated with changes in the overall proportion of DWZ. The equivalent size of the effects of the proportion and size of obstacles within the landscape on how far organisms travel makes it meaningless to discuss the effects of changing either of these variables without making reference to any possible changes in the other. The situation was identical in the case of downstream dispersal distance. Once again, the sizes of the differences among landscapes containing different proportions of obstacles approached the sizes of the differences caused by changing the size of obstacles in landscapes with a similar proportion of obstacles (Fig. 6). For example, mean downstream distance travelled through landscapes containing 5% obstacles was 129 cells when obstacles were, on average 6 × 6 cells, and only 53 when obstacles were, on average 2 × 2 cells (a decrease of 77 cells). Increasing the proportion of obstacles to 15% (but making them, on average, 6 × 6 cells instead of 2 × 2) caused a further decrease of only 10 cells (to a distance of 43 cells), and a decrease of 35 cells (to a distance of only 17 cells) when obstacles were small (an average of 2 × 2 cells).

Figure 6.

Mean total and downstream distance (± 1 SD) travelled with obstacles present, but settlement in DWZ only. Ps equals the probability of settlement once an organism is within a DWZ. Mean obstacle sizes are 2 × 2 cells (white columns) and 6 × 6 cells (hatched columns). Note that for total distance travelled, we have excluded cases in which the distance exceeded 1500 cells (see text for details).

While mean dispersal distances responded markedly to changes in landscape structure, differences between landscapes were even better illustrated by considering the frequency distribution of dispersal distances for each landscape type (Fig. 7). Restricting settlement to DWZ changed not only the mean distance dispersed, but the entire shape of the distribution, and thus the predictability of dispersal patterns among replicates of a particular landscape type. As previously mentioned, changing the probability that organisms would settle within a DWZ did not alter the overall nature of the results. However, as the probability of settlement decreased, the effects of patch configuration on dispersal distances decreased relative to the effect of the overall proportion of patch types in the landscape (Fig. 6).

Figure 7.

Frequency distributions of average downstream distance travelled per landscape, for each landscape type. (a) 5% obstacles, random settlement; (b) 15% obstacles, random settlement; (c) 5% obstacles, settlement in DWZ only; (d) 15% obstacles, with Pr(settlement within DWZ) equal to 1·00. Mean obstacle sizes are 2 × 2 cells (white columns) and 6 × 6 cells (hatched columns).


The results of this simple model show quite clearly that, first, adding patchiness to the landscape and, secondly, modifying settlement behaviour, can significantly alter the potential movement distances of dispersing organisms. The addition of obstacles (and DWZ) into streams in our model decreased the distances organisms drifted downstream (Fig. 5), principally by causing them to move laterally more often, and also by creating crevices that could trap organisms, both processes that are likely to occur in natural stream channels. When organisms did not make any distinction between patch types, and settled at random within the landscape, changing the size of obstacles had a relatively small impact on movement patterns relative to the effects associated with increasing the actual proportion of obstacles in the stream.

However, when the behaviour of organisms was modified such that they settled only in the DWZ downstream of obstacles, changing the size of obstacles (and DWZ) influenced transport distances to the same degree as changing the proportion of obstacles within a reach. This suggests that the importance of landscape structure for movement patterns is highly contingent upon whether or not organisms settle at random or only in DWZ. Consequently, unless the nature of DWZ (i.e. their size and positioning) is controlled in attempts to relate the proportion of DWZ to drift distances, then these effects may well be confounded. Because, on average, downstream distance travelled differed among landscape types even when settlement was not restricted to DWZ, we suggest that settlement in DWZ may not be a necessary condition in explaining the results of previous studies where this assumption has been made (Lancaster et al. 1996). Thus, if organisms settle at random, at a rate described by an exponential decay model, and the important factor in explaining intersite differences in dispersal distance is velocity, then measuring the proportion of DWZ may do no more than provide a surrogate method for estimating average velocities within a reach. If, on the other hand, organisms do settle only in DWZ, then a measure of the size and spatial arrangement of DWZ may be necessary to properly understand how dispersal distances might vary among different stream reaches, and also over time with changes in discharge. This could provide an explanation for the observed differences in drift distances among reaches with similar proportions of DWZ in the study of Lancaster et al. (1996). Furthermore, our results conform with those of other studies examining the effects of landscape patchiness on organism movement patterns. To take a rare example from the stream literature, Lancaster & Belyea (1997) examined the role of ‘flow refugia’ (analogous to DWZ) in regulating population responses to repeated disturbances using a simple dispersion model. They found that many small refuges harboured larger total population sizes than a few large refuges. A larger population size in their model is akin to shorter dispersal distances in ours because, in both cases, very long dispersal distances amount to death (J. Lancaster, personal communication). These results are in agreement with the results of similar studies noting the importance of patch configuration to dispersal processes in patchy terrestrial landscapes (Bergelson et al. 1993) where this issue has generally been given far more attention.

Our results also showed that, in situations where there were a few but large DWZ within reaches, travel distances were far more variable (Fig. 7). If drift distances vary among real stream reaches in the way that our results suggest, then the longitudinal connectedness of populations is likely to vary as well. In such cases, understanding temporal and spatial changes to populations may come more from focusing on this variability, rather than on average dispersal distances.

In our analysis we included scenarios in which settlement probabilities were less than one, to reflect the fact that many stream organisms are either highly mobile or sedentary rather than sessile, and therefore are potentially able to decide to move again if they initially settle in an unfavourable location. Surprisingly, however, reducing settlement probability had relatively little effect on the distances over which the organisms dispersed. The greatest effect was observed in landscapes containing many small DWZ, most probably because organisms entering a large DWZ have a chance of settling in subsequent iterations as they continue to move downstream through the DWZ (with settlement probabilities at each time-step spent in the DWZ of 0·25, 0·50 and 0·75, respectively). A further level of complexity that might prove interesting is to consider a landscape in which the quality of patches varies such that some patches have a higher number of organisms opting to settle upon entering the patch.

In summary, the relevance of our findings in understanding more about dispersal in streams will be largely contingent on testing the assumption that settlement of drifting invertebrates is restricted to DWZ. Clearly, the degree to which this assumption is justified is likely to determine the importance of the spatial structure of DWZ in influencing dispersal processes in streams. Research efforts directed towards understanding settlement cues have shown to be critical in understanding colonization patterns and population dynamics in other environments (Keough & Downes 1982; Rodriguez, Ojeda & Inestrosa 1993), and such studies are sorely needed in streams (Palmer, Allan & Butman 1996; Downes & Keough 1998). One approach, that despite its apparent simplicity we suggest should be avoided, is to try and infer settlement cues from the distribution of organisms on the bed. While many organisms show limited distributions with respect to velocity (Wetmore, Mackay & Newbury 1990; Quinn & Hickey 1994), there is no evidence that we know of to show that they arrive at these sites directly from the drift. They may first disperse over large distances within the drift and, perhaps having settled in an area of low flow, move to these high velocity areas by crawling along the bed or, alternatively, may settle at random and choose to stay only if conditions are suitable. Many stream organisms are clearly active swimmers and are unlikely to behave like passive particles. Ciborowski (1983) released live and dead Baetis spp. nymphs into the water column and found that live animals tended to settle more rapidly and travel further laterally than dead nymphs, suggesting that they have some control over where they settle, and thus do not behave as a passive particle might. On the other hand, there are other taxa (e.g. some dipterans, and perhaps also plant propagules) for which this assumption is probably quite valid (Fonseca, 1999). Thus different taxa are likely, not only to have generally different dispersal profiles, but also to respond to changes in the nature of DWZ in quite different ways. Comparative studies of dispersal and settlement behaviour in passive and active dispersers may prove profitable in this regard. To this end, understanding not just why organisms drift, but when and how they leave the drift, is likely to be pivotal in gaining a better understanding of dispersal processes in streams.

On a more general note, these results are just as likely to apply to other systems in which dispersal is dependent upon passive transport of propagules (predominantly, although perhaps not necessarily, in one direction) and in which physical transport processes are influenced by patchiness in the landscape. Although we sized our landscape to mimic a reach of river, the cell size is not explicitly defined, and thus the results are not strictly scale-restricted. They could apply to a variety of situations in quite different environments, such as larval dispersal in marine systems and seed dispersal through forests. In marine systems, the importance of settlement cues and larval supply to the distribution of adult organisms in the intertidal zone has been well documented, but physical processes affecting local larval densities are less well understood, at least at small spatial scales (Downes & Keough 1998).

Finally, our study adds support to the argument that spatially explicit processes such as dispersal can, at least in some instances, be best understood through the use of spatial models (Steinberg & Kareiva 1997). While there is a large body of theoretical literature to support this idea, empirical evidence is still somewhat lagging (Steinberg & Kareiva 1997). This paucity of evidence undoubtedly stems from the practical constraints imposed on experimental research, an issue in our own decision to adopt the current method of inquiry. This problem is, however, one that must be overcome for further advances to be made. We see several ways in which the current results might be tested in real stream environments. Patch structure can be easily manipulated in small stream reaches, and individual organisms could perhaps be simulated using small (yet easily visible) plastic pellets of the appropriate size and weight. Even if these behaved as poor mimics of real organisms, they might serve as a way of mapping the size and location of DWZ (including submerged and very small ones) within a stream reach relatively quickly and cheaply. Such methods may allow the required level of replication to be achieved without imposing the difficult task of trying to follow the fate of individual organisms, a problem afflicting dispersal studies in many systems. Flume experiments employing real organisms provide an alternative, yet equally feasible, approach to dealing with the trade-off between realistic representations of natural stream channels and the characteristics of individual organisms (Fonseca, 1999). While such abstractions have obvious limitations, they may prove valuable in making the timely link between the theoretical predictions and the empirical patterns associated with space in ecosystems.


We would like to thank Jill Lancaster, Lisa Belyea, Ashley Sparrow and two anonymous referees for their helpful comments on an earlier draft of the manuscript. This research was carried out with the help of financial support provided through an Australian Postgraduate Award scholarship to Nick Bond and a Melbourne Research Scholarship awarded to George Perry. Fieldwork costs were covered as part of an Australian Research Council grant awarded to Barbara J. Downes.

Received 4 August 1999; revision received 8 December 1999