Would conserving natural land cover in landscapes conserve biodiversity?

It is generally accepted that protecting natural land cover would protect biodiversity. This would only be true as a general statement if the relationship between richness and natural land cover were monotonic positive and scale- and method-independent. Assertions about habitat loss causing species losses often come from broad-scale assessment of richness (e.g., from range maps) combined with patterns of natural habitat conversion. Yet, the evidence about species loss following habitat loss or fragmentation typically comes from fine-scale experiments. Here, we test whether broad-extent relationships between avian species richness and natural land cover are independent of: 1) whether species distribution data come from systematic censuses (atlases) versus range maps, and 2) the grain size of the analysis. We regressed census-based and range map-based avian species richness against the proportion of natural land cover and temperature. Censused richness at the landscape level was obtained from Breeding Bird Atlases of Ontario and New York State. Range-map richness derived from BirdLife International range maps. Comparisons were made across different spatial grains: 25-km2, 100-km2, and 900-km2. Over regional extents, range-map richness relates strongly to temperature, irrespective of spatial grain. Censused species richness relates to temperature less strongly. Range-map richness is a negative function of the proportion of natural land cover, while realized richness is a peaked function. The two measures of richness are not monotonically related to each other. In conclusion, the data do not indicate that, in practice, landscapes with greater natural land cover in southern Ontario or in New York State have higher species richness. Moreover, different data types can lead to dramatically different relationships between richness and natural land cover. We argue that the argument that habitat loss is the main driver of species loss has become a panchreston. It may be misguiding conservation biology strategies by focusing on a threat that is too general to be usefully predictive.

The NYBBA sampled birds on a 5x5km grid; the OBBA used a cell size of 10x10km.
119 Both atlases used experienced birders to identify the breeding bird species occurring within each 120 quadrat. Since the sampling was designed to sample all habitats in a grid cell, and hopefully to 121 find all species breeding there, we treat species not observed in a cell as being truly absent [49].
122 Richness in a quadrat represents the total number of species presences observed in that quadrat.

123
Sampling effort varies both within and between the two atlases. For the NYBBA, 124 atlassers were assigned to survey one or more NYBBA quadrats and were expected to spend at 125 least 8h in each block, visiting each habitat present, and recording at least 76 species. For the 126 ABBO surveys, each volunteer was assigned to search a specific 100-km 2 quadrat as completely 127 as possible for evidence of all species breeding therein. Volunteers were instructed to search in 128 particular for regionally rare species. Any species that was observed in a given quadrat in 2000-129 2005 was considered present. We excluded ABBO quadrats with <20 hours and > 600 hours of 130 bird censusing effort (median effort hours). Since the OBBA quadrats were four times ≅45 131 larger than the NYBBA quadrats, the effort per unit area was similar in the two atlases.

132
In order to compare censused richness between atlases, we resampled the NYBBA at 133 10x10 km quadrat size (same cell-size as ABBO

178
The relationship between avian species richness and natural land cover depends on the type 179 of data from which richness gradients are generated. Range-map richness (Fig 3) clearly reflects 180 the climatic gradients in the region (Fig 2) rather than land cover (Fig 1). Like climate, the 181 spatial variation of range-map richness is strongly autocorrelated in space, across cell sizes from 182 25 km 2 -900 km 2 (Fig 3a-c). Multiple regressions showed that range-map richness relates 183 strongly to temperature, and it is negatively related to the proportion of natural land cover (Table   9 184 1, S3 Fig), contradicting the expected positive richness-land cover relationship. Moreover, most 185 of the small amount of variance in range-map richness explained by natural land cover (Table 1) 186 reflects collinearity between land cover and climate: forested areas in New York and Ontario are 187 mainly in cold areas (S1 Fig). The observed patterns are very similar across 25-km 2 , 100-km 2 , 188 and 900km 2 cell sizes (Fig 3a-c, Table 1).   Censused richness is only weakly related to range-map richness over this study area (Fig 6), An effect of land cover on censused richness is most apparent in multiple regression models, 235 after controlling for range-map richness (Table 3). Yet, the relationship is peaked with maximum 236 censused richness reached roughly between 52-65% natural land cover, depending on the data 237 type and grain size (Fig 5b,d,f). These results are only consistent with the proposition that 238 protecting natural cover protects richness at very low natural cover.  Spatial autoregressive models do not change the qualitative patterns described above.
244 Overall, these models performed better than the OLS models based on AIC comparisons (Tables   245 1, 2 main text vs. S1-S2 Tables). However, incorporating spatial autocorrelation increased 246 variance explained of range-map richness models by 1% (comparison of r-squares from ≤ 247 Table 1 vs. S1 Table), and the regression coefficients differ little. For censused richness, the 248 additional variance explained added by autoregressive error models was also very small, but it 249 seems to be more relevant at the 30x30km cell size (comparison of r-square from Table 2 vs. S2   250 Table).

265
Yet, when examined closely, richness generated from range maps relates negatively to the 266 proportional of natural land cover (i.e., mostly forest cover), probably because the patterns of 267 richness are mainly driven by temperature (Fig 3, Table 1). Land cover varies dramatically over 268 relatively fine spatial scales (Fig 1)