Biological Invasions

, Volume 12, Issue 12, pp 4019–4031

Introduced weed richness across altitudinal gradients in Hawai’i: humps, humans and water-energy dynamics

Authors

  • Gabi Jakobs
    • Department of BotanyUniversity of Hawai’i at Manoa
  • Christoph Kueffer
    • Department of BotanyUniversity of Hawai’i at Manoa
    • Institute of Integrative Biology, ETH Zurich
    • Department of BotanyUniversity of Hawai’i at Manoa
Original Paper

DOI: 10.1007/s10530-010-9816-6

Cite this article as:
Jakobs, G., Kueffer, C. & Daehler, C.C. Biol Invasions (2010) 12: 4019. doi:10.1007/s10530-010-9816-6
  • 183 Views

Abstract

Native species richness commonly declines with increasing altitude, but patterns of introduced species richness across altitudinal gradients have been less frequently studied. We surveyed introduced roadside weeds along altitudinal transects ranging from 30 to 4,100 m in Hawai’i, with the objectives of (1) testing the hypothesis that a mass effect due to mixing of tropical and temperate species at mid-elevation promotes a hump-shaped pattern of introduced species richness with altitude, and (2) testing the potential roles of anthropogenic activity, energy (temperature) and water-energy dynamics (productivity-diversity hypothesis) in determining introduced weed richness. A total of 178 introduced weeds were recorded. Introduced weed richness does not decline monotonically with altitude. Rather, mixing of tropical and temperate species helps to maintain high mean richness up to 2,000 m, suggesting a mass effect, but without a distinct richness peak. Patchy occurrence of a transformer species, Pennisetum clandestinum, introduced high variance in richness at mid-elevations. General linear models considering estimated actual evapotranspiration (AET, a measure of energy-water dynamics) together with an index of human activity (distance from urban area or length of major roads) accounted for more variance in introduced weed richness than models with energy alone (temperature) and human activity. Native Hawaiian species richness along roadsides was also weakly correlated with AET but negatively associated with human activity. Our observed association between introduced species richness and AET mirrors patterns reported for native species richness around the world, indicating that AET-richness patterns can develop on a short time scale (on the order of 100 years). To test the generality of introduced weed richness patterns, we tried using the Hawai’i island model to predict weed richness on the neighboring island of Maui. Although weed richness on Maui was under-predicted, the same predictors (human activity and AET) were important on Maui. Scaling for differences in regional human population density or economic activity (both higher on Maui) may allow more accurate and transferable quantitative predictions of introduced weed richness patterns.

Keywords

AlienDisturbanceEvapotranspirationHawaiiInvasiveMauiNon-nativePlantsRoadsideTropical

Introduction

Understanding spatial patterns of species richness has been a key objective of naturalists, ecologists and biogeographers for centuries (Forster 1778; Wallace 1889). A recent review of empirical literature concluded that variables associated with productivity are most often the best predictor of species richness across large spatial scales (Field et al.2009). The productivity-diversity hypothesis predicts an increase in species richness with productivity, which sometimes peaks below the highest level of productivity (Zobel and Partel 2008). In plants, measures of potential productivity based on water and energy appear to be especially valuable indicators of species richness (Hawkins et al. 2003; O’Brien 2006), helping to explain broad geographic gradients in diversity from the tropics to high latitudes (e.g. Fischer 1960). Variation in productivity may also help us understand regional, landscape and local patterns of species richness (Mittelbach et al.2001), although at smaller spatial scales other factors such as physical heterogeneity, biotic interactions and colonization limitation may become equally important (Field et al.2009). Other lines of reasoning suggest that energy alone should predict broadscale species richness patterns. Metabolic theory predicts a direct and positive relationship between species richness and temperature (Allen et al. 2002), while Rohde (1992) proposed that increasing temperatures can affect evolutionary rates and patterns, promoting increased species richness.

Patterns of species richness across altitudinal gradients have attracted particular attention, perhaps because species turnover and sharp changes in community structure are often obvious in mountain systems (Lomolino 2001). Many studies have found a monotonic decline in native species richness with increasing elevation (Stevens 1992), but a hump-shaped pattern of richness may also be observed, particularly when a wide altitudinal range has been sampled (Rahbek 1995). A hump-shaped pattern of richness versus altitude is expected if stronger environmental filters at extreme high and low elevations reduce the number of species that can survive there. Higher species richness at mid-elevations may also be predicted by random placement of species along an altitudinal gradient (i.e. the mid-domain effect; Colwell et al. 2004).

Relatively few studies have examined patterns of introduced species richness across elevation gradients (Pauchard et al. 2009), but patterns may differ from those of native species. Most introduced species are imported to lowland areas in association with human activity there, and introduced species thrive in areas of human disturbance (Hobbs and Huenneke 1992). Human activity and disturbance often decline exponentially with elevation (e.g. Leu et al.2008), and this trend may be expected to drive patterns of introduced species richness, resulting in sharp and monotonic declines in richness with increasing elevation. In fact, Becker et al. (2005) reported an exponential decline in alien species richness with elevation in the Swiss Alps. In that study, maximum altitude of species occurrence was weakly correlated (r = 0.22) with time since introduction, suggesting that dispersal limitation may have contributed to the observed pattern; alternatively, the authors suggested that insufficient time for adaptation to higher elevations may presently restrict recent introductions to lower elevation (Becker et al. 2005). A recent study in the Canary Islands confirmed this effect of residence time, whereby archeophytes introduced for >500 years reached higher elevations than more recently introduced neophytes; and this effect was particularly pronounced among non-temperate species that were not climatically pre-adapted to a high elevation climate (Haider et al. 2010). Other surveys of introduced plants along altitudinal transects in Chile (Pauchard and Alaback 2004), South Africa (Kalwij et al.2008), and temperate Australia (Mallen-Cooper and Pickering 2008; McDougall et al. 2005) have revealed linear declines in species richness with elevation, whereas hump-shaped patterns were reported for sites in the tropics (La Reunion; Tassin and Riviere 2003) and subtropics (Canary Islands; Arévalo et al. 2005). Arévalo et al. (2005) discussed the possibility that high productivity at mid-elevations could have promoted a peak in alien (and native) species richness at mid-elevations, but they did not address this hypothesis quantitatively.

The Hawaiian Islands offer an outstanding study system for analyzing patterns of introduced species richness across altitudinal gradients because of their wide range in elevation (0–4,100 m), varying climatic conditions on leeward and windward slopes (with a trade-wind inversion at ~2,000 m causing sharp climatic contrasts; Giambelluca and Schroeder 1998), and a well established road system that allows easy access to most of the altitudinal spectrum. Roadsides are a useful habitat for studying introduced species richness because associated road fill is often a relatively uniform substrate, and the road provides a consistent reference point for sampling. Roadsides have also attracted broad interest in conservation biology because they are often points of initial establishment for invasive weeds that threaten natural areas or other human interests such as agriculture (Gelbard and Belnap 2003). Most introduced plants found naturalized along roadsides are unwanted because they present various hazards ranging from flammability to obstructing viewpoints or breakdown lanes; therefore, we refer to these plants as introduced weeds. Wester and Juvik (1983) sampled roadside plots from 500 to 2,500 m on Mauna Kea, Hawai’i and reported a general decline in introduced weed richness with elevation; however, the nature of the decline was difficult to discern. Among nine sampled plots, a decline in richness is only apparent in their last two plots near 2,500 m. With increasing elevation, Wester and Juvik (1983) reported a shift in introduced weed origin, from tropical to temperate species. A shift to temperate invaders at higher elevations in Hawai’i was also documented by Daehler (2005) using floras and other information on local range limits. These observations raise the possibility that a peak in alien species richness may occur at mid-elevations where tropical and temperate introduced species co-exist. Such a pattern might be expected based on analogous situations where species richness peaks have been associated with ecotones due to a ‘mass effect’ (Shmida and Wilson 1985).

The purpose of this study was to determine the pattern of roadside alien plant richness across broad altitudinal transects in Hawai’i in order to (1) test the hypothesis that mixing of tropical and temperate species at mid-elevation climatic ecotones promotes a hump-shaped pattern of introduced species richness with altitude (mass effect; Shmida and Wilson 1985), and (2) test the potential roles of anthropogenic influence, energy (Allen et al. 2002) and water-energy dynamics (productivity-diversity hypothesis; Hawkins et al. 2003) in determining introduced weed richness. We also compare the influence of these variables on native species richness. Although direct measures of plant productivity are rarely available for geographic-scale studies, productivity is expected to correlate with actual evapotranspiration (AET), which is a function of radiation, temperature and water availability (Hawkins et al.2003). For our study, AET was estimated as an indicator of productivity while indices of human activity (anthropogenic disturbance and propagule pressure) were calculated based on distance between sample plots and the nearest urban area and length of major roads at different elevation bands.

Methods

Study area and field survey

Surveys were done between March and July 2007. To investigate introduced (non-native) species distributions along altitudinal gradients in the Hawaiian Islands, we surveyed four altitudinal transects on the three tallest Hawaiian volcanoes: Mauna Kea (19°45′N, 155°, 27′W; island of Hawai’i, two transects), Mauna Loa (19°29′N, 155°36′W; island of Hawai’i) and Haleakala (20°57′N, 157°16′W, island of Maui). All transects were located along major roads (Fig. 1), although vehicular traffic generally decreases with increasing elevation. Elevations ranged from 30 to 4,100 m, depending on the transect.
https://static-content.springer.com/image/art%3A10.1007%2Fs10530-010-9816-6/MediaObjects/10530_2010_9816_Fig1_HTML.gif
Fig. 1

Map showing locations of roadside transects on the islands of Hawai’i and Maui. Surveys were made of 50 × 2 m plots along roads at approximately 150 m elevation intervals

Surveys were made at roughly 150 m elevation intervals along the four transects. At each survey site, a 50 × 2 m plot was established parallel to and abutting the road, but excluding any bare gravel zone, if it existed. Left or right sides of the road were chosen at random. Plots were initially divided into 10 × 2 m segments and species accumulation curves showed a general leveling of species richness after pooling of five such plots, demonstrating appropriateness of the plot size (results not shown). All introduced and native angiosperms in the plots were recorded. The introduced angiosperms occurring in these disturbed roadside plots are collectively referred to as introduced weeds.

Environmental and species data

In order to test the hypotheses that mixing of tropical and temperate zone species promotes a hump-shaped pattern in richness (mass effect hypothesis), the native ranges of all species were broadly classified as either primarily tropical/subtropical (within 23.5°N or S latitude) or primarily temperate (between 23.5 and 66.5°N or S latitude; Tarr and McMurry 1919) based on native range information listed in USDA GRIN taxonomy data base (http://www.ars-grin.gov/). To test the hypothesis that introduced weed richness will be associated with the degree of anthropogenic influence, two indices of human activity were calculated in the vicinity of each plot: major road lengths associated with 300 m elevational bands around each plot, and proximity of plots to urban areas. These indices were determined from GIS layers for major roads and urban areas, including low density urban areas and resort areas (Hawaii Statewide GIS program, http://hawaii.gov/dbedt/gis/). Major roads, urban and resort areas were defined and delineated by the County of Hawai’i Planning Department (Hilo, Hawai’i). Length of major roads was used as predictor because it is a measure of human activity and habitat area available for roadside weeds. For plots above 150 m asl., elevation bands were centered around the plot elevation; for the two lower plots, the band from sea level to 300 m was used. Road length was used as predictor rather than road density (standardization by total land area in each elevational belt) because most weeds recorded in our surveys were restricted to roadsides or other highly disturbed areas.

To test for the predicted relationship between energy and species richness, mean annual temperature was used as a measure of energy (Allen et al. 2002). Climate data (1971–2000 normals for mean monthly temperature as well as monthly rainfall) were obtained from the Prism Climate Group (http://www.prism.oregonstate.edu/) at a resolution of 15 arcseconds. To test for the hypothesized relationship between productivity and richness, we used an estimate of AET. To estimate AET, solar radiation was determined from GIS maps available from the Hawai’i State Department of Planning and Economic Development, Energy Division (http://hawaii.gov/dbedt/gis/data/solrad.txt) Potential evapotranspiration (PET) was then estimated using a Hargreaves-Samani model (Hargreaves and Samani 1982), which had been calibrated for Hawai’i (Wu 1997). Values for monthly PET were calculated based on monthly precipitation, monthly temperature, and solar radiation, then summed across the year. Actual evapotranspiration (AET) was then estimated by scaling PET by an aridity index (AI), defined as P/PET where P is precipitation (Middleton and Thomas 1992), with the maximum allowable AI being one. The resulting estimate for AET then becomes the lesser of monthly precipitation or PET for each month, summed over all months of the year. Both AET and PET were tested for their relationship to richness, since PET is sometimes used as an indicator of productivity, although AET is usually preferred (Whittaker and Heegaard 2003).

Statistical analyses

Most statistical analyses were conducted using Systat version 13 (Systat Software, Inc). The relationship between species richness and elevation was assessed by comparing the fit of linear versus a quadratic (hump-shaped) regression models using data from the island of Hawai’i. Separate regressions were then made for temperate and tropical species to identify the influence of each group on the overall pattern. General linear models (GLMs) were used to determine the relationship between species richness and the potential explanatory variables, which were transformed when necessary to accommodate assumptions of normality and homogeneity of variance (transformations given in parentheses): mean annual temperature, PET (inverse), mean monthly precipitation, AET (square root), length of major roads (natural log-transformed + 1), and distance to nearest urban area (square-root). An AET-squared term was also included to test for a hump-shaped relation between productivity and species richness, as has sometimes been reported for native species (Mittelbach et al. 2001). GLMs assuming a normal distribution were used because the count distribution of introduced weed species observed in our plots was reasonably approximated by the normal distribution, whereas the Poisson was a poorer fit (variance ≫ mean) and a negative binomial distribution was rejected (chi-square = 14.5, df = 6, P = 0.025). Best fit GLMs were identified by comparing results of forward and backward stepwise inclusion of the explanatory variables, with the criterion for variables to enter or be removed from the models set at P ≤ 0.10. The resulting models were compared using the Akaike information criterion (AIC). To test for potential influence of spatial autocorrelation on regression model parameters, the first order spatial trend surface (based on physical distance between plots) was included together with water-energy and anthropogenic factors using SAM software (Rangel et al. 2006). To test the generality and transferability of the best fitting GLM model, predictions of weed richness were made for Maui and compared to actual weed richness along the Maui transect. Relationships between native species richness and explanatory variables were also tested for comparison to patterns identified for introduced weeds. For GLMs based on native species counts, a negative binomial distribution was used based on adequate fitting of the native species counts (no significant departure from negative binomial; chi-square = 1.63, df = 1, P = 0.202). Negative binomial GLMs were performed using the R statistical package version 2.9.0 (R Development Core Team 2009).

Results

Introduced weed richness versus elevation

A total of 178 introduced weeds were recorded, of which 55 were primarily temperate in origin, 118 were tropical, and five were unclassified. On the island of Hawai’i, each of the three transects had somewhat different elevational ranges due to limitations of roads and differing maximum heights of volcanoes (Hilo-Mauna Kea, 290–4,100 m; Kona-Mauna Kea, 30–3,200 m; Mauna Loa, 500–1,800 m). Species richness values overlapped substantially among transects when similar elevations are compared (Fig. 2). Considering elevational ranges of overlap between transects, the slopes of best-fit linear regression lines were not statistically different between Hilo and Kona transects on Mauna Kea (P = 0.45, t = 0.77, df = 38) or between Kona and Volcano (P = 0.27, t = 1.13, df = 16); therefore, the three transects were pooled for overall analyses of introduced weed richness patterns on the island of Hawai’i.
https://static-content.springer.com/image/art%3A10.1007%2Fs10530-010-9816-6/MediaObjects/10530_2010_9816_Fig2_HTML.gif
Fig. 2

Non-native plant species richness along roadsides versus elevation on the island of Hawai’i. The best-fit quadratic regression pooled across transects is hump-shaped (solid line). The dashed regression line represents fit of the data following exclusion for the four lowest elevation plots on the Mauna Loa transect (indicated by asterisks), which occurred on roadside bordering relatively young lava. Transects are indicated by letters: Hilo-Mauna Kea (H), Kona-Mauna Kea (K) and Mauna Loa (M)

A linear regression model provides a weaker fit of the richness-elevation pattern (r2 = 0.392, slope = −0.0045, y-intercept = 21), relative to a quadratic regression model (r2 = 0.433, equation given on Fig. 2). The improvement in fit of the quadratic regression over the linear regression was statistically significant (P = 0.02, F2,55 = 3.98). Four lower plots on the Mauna Loa transect occurred along roadside bordering relatively young lava (70–150 years old), possibly making their richness values lower than the other plots, and influencing the shape of the richness-elevation relationship (Fig. 2). When these four plots are excluded, the quadratic regression model remains a better fit of the data than a linear model (P = 0.03, F2,51 = 3.76). The mathematical peak of the regression is at 385 m or 286 m for the full data set or the reduced dataset, respectively; however the peak is extremely broad in the range of 0–1,000 m, with the projected richness differing by <2 species within this range. The evidence for hump-shaped pattern of richness versus elevation is therefore quite weak, and the data may be fit equally well by a split-function regression, with a linear decline in species richness above 2,000 m and no pattern (zero slope with a high mean richness and variance) below ca. 2,000 m (see Fig. 2).

To determine whether overlap of temperate and tropical zone species influences the overall shape of the species richness—elevation relationship, we fit a quadratic model to the data after excluding the temperate zone species (Fig. 3). The resulting best fit regression line is no longer a hump-shaped pattern but is instead concave, consistent with an exponential decline in tropical weed richness with elevation, and indicating that temperate zone species play an important role in increasing species richness at mid-elevations, thereby altering the shape of the richness curve from concave (tropical species) to convex (all species). Considering temperate species alone, richness peaks at around 2000 m elevation (Fig. 3). Although a quadratic fit for temperate species across the full elevation range is significant, the pattern above 2,000 m may be better characterized by a linear or exponential decline (Fig. 3).
https://static-content.springer.com/image/art%3A10.1007%2Fs10530-010-9816-6/MediaObjects/10530_2010_9816_Fig3_HTML.gif
Fig. 3

Introduced weed species richness versus elevation, separated by origin of the species. The solid regression line is for species of primarily tropical origin. The dashed regression line is for species of primarily temperate origin

Explaining patterns of introduced weed richness

Mean monthly temperature, PET, and AET were not strongly correlated (all r < 0.67), whereas the indices of human disturbances and propagule pressure (length of major roads and distance from urban areas) were more strongly correlated (r = 0.88) and both were negatively correlated with temperature (Table 1), reflecting the expected decreasing influence of humans with increasing elevation and cooler temperatures (Fig. 4). Stepwise model fitting identified a model containing distance from the nearest urban area together with AET as the best fit model accounting for 50% for the observed variation in introduced weed species richness (radj2 = 0.50), with distance from urban area having a negative influence (t = 4.56, P < 0.001) and AET having a positive influence (t = 2.95, P = 0.005); richness = 8.066 *Sqrt(AET)−1.805* Sqrt(Distance from urban area) + 9.124. The condition number (CN) for this model, after standardizing the variables, was 1.67, suggesting that multicollinearity had little influence on the model parameters (Lazaridis 2007). Other models tested had higher AIC values (Table 2), although a model replacing distance from urban area with length of major roads also had a relatively low AIC, as would be expected due to correlation between these variables (Table 2). Also, models substituting temperature and precipitation for AET had only slightly higher AIC (Table 2), as may be expected because AET is a function of temperature and precipitation. Including spatial structure in the regression model increased the fit of the model (radj2 = 0.649), with 10.5% of the variance explained by spatial structure and 4.7% of the variance shared between spatial structure and the predictor variables. Both AET and distance to urban area remained highly significant (P < 0.001) after incorporation of spatial structure, indicating that the association of these variables with weed species richness is not accounted for by spatial autocorrelation. A quadratic AET term was not significant, and adding a quadratic term results in a weaker fitting model (Table 2), indicating lack of evidence for a hump-shaped relationship between richness and productivity. Models considering temperature alone or in combination with human activity were a weaker fit than those involving AET (Table 2).
Table 1

Pearson correlation coefficients between explanatory variables used in multiple regression models for the island of Hawai’i

 

Mean temperature

Potential evapotranspiration

Actual evapotranspiration (sqrt)

Road length (ln)

Distance from urban area (sqrt)

Potential evapotranspiration

0.662

    

Actual evapotranspiration (sqrt)

0.474

−0.193

   

Road length (ln)

0.969

0.514

0.536

  

Distance from urban area (sqrt)

−0.912

−0.497

−0.488

−0.877

 

Annual precipitation (sqrt)

0.525

−0.134

0.810

0.551

−0.566

https://static-content.springer.com/image/art%3A10.1007%2Fs10530-010-9816-6/MediaObjects/10530_2010_9816_Fig4_HTML.gif
Fig. 4

Length of major roads, an index of human activity, decreases exponentially with increasing elevation

Table 2

Comparison of models fitting introduced weed species richness on the island of Hawai’i to various explanatory factors

Model typea

Model

AIC

AIC excluding lower Mauna Loa

W-E, A

AET + distance from urban

362.9

331.1

W-E, A

Temp + precip + distance from urban

363.3

 

W-E, A

AET + roads

365.9

324.7

W-E, A

Temp + precip + roads

 

327.9

E, A

Temperature−1 + distance from urban

366.1

339.1

E, A

Temperature−1 + roads

369.7

336.5

A

Distance from urban

369.4

345.1

E

Temperature−1

373.2

344.4

W-E

AET

379.5

345.9

W-E

AET + AET2

380.2

347.6

A lower Akaike information criterion (AIC) statistic indicates a better fitting model. Models were fit using data pooled from all transects. AIC values are also shown following exclusion of the four lower sample plots on the Mauna Loa transect (see Fig. 2). Bold indicates the best fitting model based on AIC

aW-E water-energy, E energy, A anthropogenic activity

Models were also fit after excluding the four lower plots on the Mauna Loa transect. Exclusion of these plots seemed most appropriate for testing transferability of the model, since other islands lack young lava flows. In the resulting best fit model after forward and reverse step-wise model selection, AET and length of roads both had positive effect on richness (Fig. 5; t = 3.67, P = 0.001 and t = 5.23, P < 0.001, respectively; richness = 9.849*Sqrt(AET) + 2.085*Ln(Roads + 1)−7.555, radj2 = 0.61). Other models tested had higher AIC (Table 2). When the best fit model for Hawai’i was tested for fit to the Maui survey data, the model strongly under-predicted richness on Maui (Fig. 6; see also Fig. 5). However, refitting coefficients for AET and length of roads to the Maui data led to a model that accounted for a greater proportion of the variance in species richness (92%) than the best fitting model for the island of Hawai’i (50%). AET and anthropogenic indicators were more strongly correlated on Maui, making it difficult to separate their potential contributions to introduced weed richness. In single variable regression models, AET accounted for 79% for the variance in introduced weed richness while distance from the nearest urban area accounted for 76% of the variance. Correlation between these variables was −0.78 for Maui, versus −0.49 for Hawai’i.
https://static-content.springer.com/image/art%3A10.1007%2Fs10530-010-9816-6/MediaObjects/10530_2010_9816_Fig5_HTML.gif
Fig. 5

Introduced weed richness versus AET and length of roads. Solid circles represent plots from the island of Hawai’i while hollow circles are from Maui. Open triangles are four plots bordering younger lava on the island of Hawai’i that were excluded in some analyses

https://static-content.springer.com/image/art%3A10.1007%2Fs10530-010-9816-6/MediaObjects/10530_2010_9816_Fig6_HTML.gif
Fig. 6

Actual introduced weed richness in survey plots on Maui versus predicted weed species richness based on the model obtained from the island of Hawai’i that includes length of major roads (an index of propagule pressure and disturbance) and AET as explanatory factors. Points on the solid line indicate exact correspondence between the model and observed richness on Maui

Patterns of native species richness

In comparison to the richness of introduced weeds, relatively few native species were observed in our roadside plots (maximum of 6 species). Native species richness was not correlated with elevation (r = 0.023, P = 0.87). The negative binomial model coefficient for AET was positive (z = 2.67, P = 0.008) and the coefficient for length of roads (anthropogenic disturbance) was negative (z = 2.32, P = 0.02); nevertheless, a good deal of variance in native species richness is not accounted for by this model (Fig. 7).
https://static-content.springer.com/image/art%3A10.1007%2Fs10530-010-9816-6/MediaObjects/10530_2010_9816_Fig7_HTML.gif
Fig. 7

Native species richness versus estimated actual evapotranspiration (AET) and anthropogenic activity (length of major roads). At a given level of anthropogenic activity, richness tends to be higher at higher AET

Discussion

Richness-elevation shape and variance

In Hawai’i, introduced weed species richness does not appear to be an exponentially or monotonically decreasing function of altitude, as would be predicted if human activities were the sole cause of the pattern. We hypothesized that a zone of overlap between tropical and temperate species at mid-elevations in Hawai’i would cause a hump-shaped richness pattern. Richness peaks are often associated with the intersection of many range margins (Stevens 1992), as may be common when altitudinal transects are characterized by distinct habitat zones (see also Shmida and Wilson 1985; Grytnes 2003). In fact, introduced weed richness is high but variable in Hawai’i up to 2000 m, with a decline in richness a higher elevations. Although a significant quadratic term suggested a broad humped pattern of richness, there is no clear peak in richness. Nevertheless, by separating temperate and tropical species, we found that high richness at mid-elevations (1,000–2,000 m) is associated with an environment shared by many temperate and tropical species, broadly supporting the hypothesis of a mass effect (Shmida and Wilson 1985) or the possibility of rescue effects, wherein species are recorded above and below their ranges of long-term persistence. Factors that limit the persistence of most temperate weeds at lower elevations are not well-studied but may include competition from tropical species, diseases, or physiological incompatibility with a year-round warm environment (Daehler 2005). The same general pattern involving mixing of tropical and temperate species at mid-elevations can be seen in elevation transect surveys made on the island of La Réunion (Tassin and Riviere 2003), which at 21°S latitude is the most comparable site to Hawai’i that has been surveyed for introduced weed richness.

Even without considering the possibility of a mass effect, extreme conditions at both ends of any gradient may cause a hump-shaped pattern of species richness. Rahbek (2005) emphasized that patterns of species richness across altitudinal gradients may depend on the extent of the altitudinal gradient. The upper end of the gradient in Hawai’i, at above 4,000 m, is clearly a stressful environment due to cold temperatures, high wind and solar radiation, and low water availability during much of the year. The latter factor results in low AET at very high elevations. On the other hand, at extreme low elevations near sea level, salt spray is a well-known stressor. Our lowest survey point, at 30 m, was above the zone of persistent salt spray, although some salt influence was possible. Had we surveyed literally down to sea level, a stronger hump-shaped pattern of richness versus elevation might be expected. On the other hand, if sea level sites were located in anthropogenic habitats such as around docks or urban harbors, then high propagule pressure could lead to very high richness at sea level. Therefore, for introduced species, it is not only the magnitude of the elevation gradient (scale) that may determine richness—altitude patterns (Nogues-Bravo et al. 2008), but the choice of sampling site(s) at the lowest elevation extremes will additionally be important.

We did not explore the possibility that richness-altitude patterns were caused by geometric constraints of species ranges (i.e. mid-domain effects; Colwell et al. 2004) because we had identified a specific ecological hypothesis (tropical-temperate overlap) that could generate a humped pattern. The observed separation of tropical and temperate species along our elevation transects indicate that overall patterns are not appropriately simulated by random placement of species along the gradient, as is typically done to identify expectations for a mid-domain effect. Furthermore, Currie and Kerr (2008) reviewed evidence for the mid-domain effect along environmental gradients and concluded that empirical data departed significantly from mid-domain effect predictions in 50 out of 53 cases.

Below 2,000 m, variance in species richness was high, and this pattern is strikingly similar to that observed for La Réunion (Tassin and Riviere 2003). Several plots with unusually low richness on the Kona-Mauna Kea transect were dominated by the invasive grass, Pennisetum clandestinum, suggesting that it behaves as a transformer species (Richardson et al.2000) by fundamentally altering patterns of richness even among communities dominated by non-native weeds. This species also forms nearly 100% cover at some sites on the Mauna Loa transect between 1,100 and 1,600 m (Daehler and Angelo, unpublished data), although those sites were not sampled in the present survey. It forms dense vegetative stands in natural areas and along roadsides. High tiller densities were observed in areas with and without mowing, reflecting P. clandestinum’s ability to retain cover, even when closely cropped. Thus, a transformer species can introduce high variability into species richness patterns, which may help explain why models including water-energy dynamics and anthropogenic disturbance only accounted for ~50% of the variance in introduced weed richness. Adding spatial autocorrelation to the models explained at least an additional 10% of the variance. This unattributed spatial autocorrelation might be due to historical trends in introduction sites and dispersal limitation, or it could be due to other factors not explicitly included in the models such as spatial variation in neighboring land usage or substrate. By excluding the lower plots bordering young lava on the Mauna Loa transect, fit of the models was improved, indicating the potential importance of substrate, despite the fact that most of our roadside plots were located on relatively uniform road fill that bordered the roads.

Influence of energy versus water-energy dynamics

Rohde (1992) argued that species richness along latitudinal and altitudinal gradients is correlated with energy, which promotes faster evolution and speciation (shorter generation time, higher mutation rate, and accelerated selection). The observed patterns in Hawai’i must have developed largely within the last 100 years and cannot be explained by faster post-introduction speciation at lower elevations. Metabolic theory also predicts a relationship between energy (inverse of temperature) and species richness (Allen et al. 2002), but the fit of the energy models was not as good as models that considered water-energy dynamics. Hawkins et al. (2003) reviewed studies of native species richness along environmental gradients around the world and concluded that water-energy dynamics, which are associated with productivity (Field et al. 2009), can provide a near-universal explanation for biodiversity gradients, but that energy alone is likely to be most important at extreme latitudes. In the Italian Alps, Marini et al. (2009) reported a correlation between naturalized species richness and temperature, the latter variable being highly correlated with elevation. Temperature alone is usually not a good indicator of productivity (Whittaker and Heegaard 2003), but across a gradient with relatively low variability in rainfall, temperature could correlate strongly with productivity, especially in temperate regions where temperature can limit the length of the growing season. Thus, tropical mountains can provide better environments for differentiating the effects of energy versus water-energy dynamics on species richness patterns. The observed pattern in Hawai’i might be linked to higher numbers of introductions adapted to higher rather than lower AET conditions, but some degree of post-introduction sorting and assembly must be responsible for the pattern. The roadside weeds surveyed in this study were mainly fast-growing herbaceous species that are readily dispersed through anthropogenic activities, facilitating the sorting of species and generation of species richness patterns along gradients. Arteaga et al. (2009) reported sorting of introduced tropical, Mediterranean and other temperate zone species across roadside altitudinal transects in the Canary Islands, suggesting post-introduction sorting to roadside areas that are roughly similar to their native range climatic zones; however, they did not quantitatively test relationships between richness and AET. In natural or semi-natural habitats, associations between introduced species richness and AET may also develop, but reduced immigration rates and slower species establishment in natural habitats may substantially slow this process and lead to patterns that are linked to historical factors such as locations of original plantings. A weak but significant positive relationship between AET and native species richness was detected in our roadside plots. Most native Hawaiian species are perennials with low tolerance of anthropogenic disturbances that occur along major roads. As a result, we had little power to test patterns in native species.

AET is expected to correlate with productivity, and the relation between productivity and species richness is often hump-shaped (Mittelbach et al. 2001; but see Whittaker and Heegaard 2003), with diversity declining at the highest levels of productivity. One explanation for richness decline with very high productivity is that a few competitively dominant species may thrive in highly productive environments, crowding out many competitors. Across different levels of anthropogenic activity, there is no prevailing humped pattern or leveling of richness at high AET (Fig. 6), and this is confirmed by the weaker fit of the model containing an AET-squared term, compared to AET alone (Table 2). Yet AET is positively correlated with plant cover (r = 0.40, P = 0.001), suggesting increasing potential for competitive effects at high AET. It is likely that frequent generation of open space at our roadside study sites reduced the potential for competitive exclusion (not withstanding observations on P. clandestinum cited earlier, which occurred at moderate AET rather than at the highest AET sites). Additionally, richness of aliens at all sites may be below the levels typical of continental regions, where most humped patterns have been observed; so saturation may not have occurred anywhere along the AET gradient. It should also be kept in mind that AET is an imperfect measure of productivity. For example, young lava substrates may pose harsh or nutrient poor environments that reduce productivity below what may be expected based on AET. Ideally, diversity-productivity relationships would be tested through direct measurements of biomass growth and turnover in the field (Whittaker and Heegaard 2003).

Predicting weed richness: qualitative, not quantitative success

Introduced weed richness on Maui was higher than projected from our model based on the island of Hawai’i. This suggests low transferability of quantitative predictions, although variables found to be important for the island of Hawai’i were also important for Maui. The human population density on Maui is 4.5-fold higher than that of Hawai’i (63 vs. 14 people km−2, respectively), and this could be one factor promoting higher than expected weed richness on Maui, along with greater numbers of tourist visitors to Maui (e.g. in 2007, 2.4 million visitors to Maui vs. 1.6 million visitors to Hawai’i (Anonymous 2007), which amounts to 1,277 vs. 153 visitors km−2, respectively). Population density (Denslow et al.2009) and degree of human development (Kueffer et al. 2010) have been associated with alien plant species richness at the whole island scale. In fact, the island-wide density of naturalized and casual species richness on Maui is 5.5-fold greater than on Hawai’i (0.705 vs. 0.128 species per km2). Our analyses considered distance to urban areas as a variable affecting introduced weed richness in our plots, but this does not consider other potentially important relations such as size of the nearest urban area or human population density. Adding an additional variable or scaling factor to account for population density differences and/or “luxury effects” related to economic activity (Hope et al. 2003) may be a simple way to increase our ability to predict localized weed richness patterns across islands or regions.

Conclusions

Invasions along environmental gradients such as those found on altitudinal transects provide useful systems for understanding broad ecological patterns and for testing ecological theory (Pauchard et al. 2009). Introduced weed species in Hawai’i displayed broad climatic matching to their native ranges, with tropical species dominating at low elevations and temperate zone species dominating at higher elevation. These findings provide hope that climate niche models may be useful tools for predicting weed species distributions in Hawai’i and elsewhere where similar patterns have been recorded (e.g. Arteaga et al. 2009; Tassin and Riviere 2003; but see Haider et al. 2010). Introduced weed richness appears to be linked not only to anthropogenic disturbance and/or propagule pressure, as has been well documented by previous studies (Colautti et al. 2006; Hobbs and Huenneke 1992), but also to water-energy dynamics. Nevertheless, the presence of introduced transformer species, such as Pennisetum clandestinum in our study, are likely to cause outliers of lower richness than expected based on local water-energy dynamics and anthropogenic activity. Our work suggests that monitoring of introduced weed richness along roadsides and examining the composition of plots that have lower than expected richness may facilitate early identification of some serious weeds, which may allow them to be expeditiously targeted for eradication or control.

Acknowledgments

We thank David Benitez, who helped identify species in the Volcano National Park and Saddle Road. Survey permits were facilitated by Rhonda Loh (Hawai’i Volcanoes National Park), Steve Bergfeld (Hawai’i district, Division of Forestry and Wildlife [DOFAW]), Stephanie Nagata (Office of Mauna Kea Management), Pattie Johnson (Parker Ranch), and Liz Gordon (Haleakala National Park). Lloyd Loope and an anonymous reviewer provided helpful comments on an earlier version of this paper. This research was supported by National Research Initiative Grant no. 2006-35320-17360 from the USDA National Institute of Food and Agriculture Biology of Weedy and Invasive Species Program.

Copyright information

© Springer Science+Business Media B.V. 2010