Technical Comments

Species Abundances Across Spatial Scales

See allHide authors and affiliations

Science  26 Mar 1999:
Vol. 283, Issue 5410, pp. 1979
DOI: 10.1126/science.283.5410.1979a

William E. Kunin (1) investigated the effect of spatial scale on the abundance of scarce British plant species and concluded by asking whether his observed patterns hold across a wider range of species and scales, which would permit extrapolation of scale-area curves to estimate abundance at scales that would be otherwise difficult to study. I have carried out a similar study with birds in the southern Iberian Peninsula to determine whether these patterns have wider application. I have used measurements of the abundance and distribution patterns of all breeding bird species found in five 100 × 100 km squares (2) that span a range of bioclimatic zones (3) from Portugal to southeastern Spain (4).

I tested the predictive power of coarse- and medium-scale abundance to fine-scale abundance by Kunin's linear extrapolation method (1) with the resultant regression equation of observed value = −0.675 + 0.889 ∗ predicted, F = 1026.986, R 2 = 0.882. The observed values were most usually lower than predicted by a 1:1 relationship, as was also found by Kunin (1). To test the predictive power of the scale-area curve, I then divided the species data set into one of 69 species to estimate by linear regression the area occupied at fine scale as a function of medium- and coarse-scale areas. I used the resultant relationship to predict the fine-scale abundance of the second species set (70 species). The regression model was highly significant [log (A f) = −0.653 + 1.264 ∗ log (A m) − 0.135 ∗ log (A c); R 2 = 0.899]. This model was an accurate predictor of 1 × 1 km abundance of the second subset, explaining 0.895 of the variance. Medium-scale abundance was a better predictor on its own than coarse-scale abundance (5). I conducted a series of linear regressions, leaving out one species in each case and using the resultant equation to estimate its fine-scale abundance. The coarse-scale abundance did not make a significant contribution to the predictions, and its inclusion did not reduce the variance explained by the predictions (6).

Finally, I attempted to test the value of these models on a fourth, finest scale, dataset. This was the original dataset collected at the 1-ha sampling points (2). The data differed from the others in that it estimated density (breeding pairs per hectare) and not spatial distribution. I asked the question, to what degree can spatial data at different scales predict very fine-scale abundance? Once more, I separated the dataset into two and tested the predictive power of the regression model. The first regression model was highly significant (R 2 = 0.888). The model was a very accurate predictor of abundance of the second subset at the 1-ha scale, explaining 0.917 of the variance (Fig. 1) (7).

Figure 1

Relationship between predicted abundance at 1-ha scale for a subset of 70 species of birds on the basis of a regression model that used a second subset of 69 species. Regression equation is based on fine-, medium- and coarse-scale abundance data. Model equation is log (abundance) = 3.243 + 1.632 ∗ log(A f) − 0.742 ∗ log(A m) − 0.437 ∗ log(A c); R 2 = 0.888.

I therefore conclude that the scale-area relationship proposed by Kunin (1) appears to have wider taxonomic application, although the predictive power of the models would not appear to be significantly modified, in this case, by the addition of coarse-scale data. The ability to be able to predict very fine-scale (1-ha) abundance of breeding birds from coarser data (1-km2 and 100-km2 spatial distribution patterns, and perhaps 10000 km2 to a lesser degree) opens a window for the use of the now plentiful bird atlas data, which are commonly within the spatial resolution that we have defined here as medium scale.


Navigate This Article