Our findings suggest a spatial bias on the sampling sites of small rodents in the Atlantic Forest. The well-sampled sites are close to urban centers, roads, large fragments and with greater habitat (forest) cover. These results also indicate a sampling gap in small fragments with low forest cover and sites far from cities and roads, precluding the inferences of relationship between fragmentation and small mammal biodiversity on broad-scale perspectives. In addition, based on data available on digital platforms the spatial aggregation of sampling efforts generated a 99% of spatial gap in the knowledge of small mammals. The well-sampled sites cover only 0.03% of the biome’s extension. Our results demonstrate that the small rodent's knowledge in a broad-scale perspective is spatially limited which impairs inferences about the factors that govern and change the spatial distribution, a key factor to propose conservation tasks.
Although the aim of digital accessible platforms is to provide more data on biodiversity to increase the biodiversity knowledge (Devictor and Bensaude-Vincent 2016) and helping to fill some gaps, the current available data is still limited (Hortal et al. 2007). The limitation of biodiversity information occurs because the available data were collected in different ways, by different researchers and with different purposes (Wüest et al. 2020). The imprecision in the Atlantic Forest's small rodent data can be explained by the lack of taxonomic expertise in identifying specimens in the field, by the retention of biodiversity data in museums or undigitized private collections and by the temporal degradation of information (Ladle and Hortal 2013; Meyer et al. 2015; Tessarolo et al. 2017). Additionally, there are imprecisions on species spatial information. The lack of geographic coordinates or their imprecision contributes to the increase of biogeographic ignorance. This imprecision makes part of the information “unusable” for some analyzes (Hortal et al. 2008) such as inventory completeness. Therefore, digitally accessible data, but with inaccurate information, also increases the biogeographic ignorance (Tessarolo et al. 2021) instead of decreasing them. It would be interesting if there was a validation, monitoring and mapping of biodiversity digital information so that it becomes accurate, useful and reliable. The correct digitization of biodiversity information would allow a better use of available data, cost reduction and promotion of new analyzes and/or research/inventories increasing the possibility of macroecological pattern inferences and conservation efforts based on the relationship between fragmentation and biodiversity loss.
Our findings also suggest a spatial bias from well-sampled sites to more accessible regions. The proximity to access routes and urban centers are well-known factors in generating bias in the biodiversity knowledge from different taxonomic groups (Almeida et al. 2021; Correia et al. 2019; Oliveira et al. 2016). The historic natural exploitation of the Atlantic Forest resources causes forest loss resulting in only 28% in forest original remnants (Rezende et al. 2018). The modification in the Atlantic Forest landscape resulted in a fragmented landscape with a high density of access routes and many cities (Lapola et al. 2014; Oliveira et al. 2016), including also the Protected Areas that are easily accessible by roads. The Atlantic Forest protected areas are spatially joint embracing the largests and conserved forest remnants that being easily accessible to biodiversity sampling results on broad-scale spatial biases on both accessibility and landscape perspectives. In addition, historically, the promotion of biological research in the Atlantic Forest has favored projects on a local scale and with long-term sampling (for example PELDs). These projects were able to clarify and describe diverse ecological and biodiversity patterns on a local scale, however failed to clarify patterns on a broad-scale which can also be explained by the recent shift in the macroecology view (hierarchical approaches - Guisan and Rahbek (2011), Sobral-Souza et al. (2021b). There are still a density of sites with a distance greater than 15 km from urban centers and access routes that have not been sampled, becoming priority sites for future sampling efforts.
The disadvantage of biodiversity sampling close to roads and cities is that it can be underestimated mainly because these sites may not represent regional biodiversity (Benítez-López et al. 2010; Laurance et al. 2009) and may have the effect of biological invasion (Hobbs et al. 2009). Access routes facilitate habitat degradation (Freitas et al. 2010) and for small rodent changes on landscape configuration such as edge effect, change the species composition (Rosa et al. 2018). On the other hand, sampling sites close to large centers and roads offers better infrastructure, more qualified human resources and reduces costs for researchers with few financial support (Lessa et al. 2019; Meyer et al. 2015).
We also found bias for larger fragments and with greater habitat (forest) cover. These biases may be explained by the access facility, the greater environmental heterogeneity of larger fragments and species-area ratio which are more likely to record a greater number of species in a few sampling times (Sobral-Souza et al. 2021a). Additionally, in the Atlantic Forest, larger fragments are more temporally stable and have less effects caused by human-induced landscape modification (Hansen et al. 2020). The human colonization and habitat fragmentation result in a spatial configuration where the main forest remnants are on sites closed to high human density (Ribeiro et al. 2009). Thus the largest forest fragments are spatially joint explaining the spatial aggregation pattern of well-sampled sites. In the Atlantic Forest the main forest fragments are also protected areas generating long term habitat stability and providing long-lasting, temporal samplings with little human action.
The Serra do Mar is the most conserved biogeographic region (Ribeiro et al. 2009) with 35% of Atlantic Forest well-sampled sites within this region, covering 0.15% of the spatial surface of this region. The Serra do Mar are located in the main large protected forest massifs (such as Serra do Mar State Park and Carlos Botelho State Park), however, it is a geographically extensive region that has large urban centers, such as São Paulo city. Our findings highlight that the Floresta do Interior is six times larger than the Serra do Mar and the biodiversity knowledge is also smaller, with 0.02% of surface well studied. In the Floresta do Interior, the sampling sites are spatially joint to the Serra do Mar bioregion, probably due to the presence of large urban centers and researchers Institutes (Almeida et al. 2021; Candelária et al. 2021; Lessa et al. 2019). The biogeographic regions of the northeast are practically unknown, especially São Francisco, which does not have well-sampled sites. These northeast regions are classified by the degree of endemism, which attracts taxonomists to discover new species (Silva-Soares et al. 2021) but even so areas with high rates of endemism are poorly sampled, such as Diamantina, Bahia and Pernambuco. This can be explained by the low number of universities and Protected areas in the region. The northeast is also known as an ecotone, having a vegetational mosaic with Caatinga and Cerrado enclaves which can increase the complexity to sampling Atlantic Forest taxa.
Biased biodiversity data also hamper the building of species distribution maps because the maps will reflect the sampling bias and not the true species distribution (Rocchini et al. 2011; Ladle and Hortal 2013). Landscape configuration also affects species dispersion (Fahrig 2005) and not knowing the true species distribution based on their sampling bias may be to impairs inferences about the effects of landscape modifications on local and regional biodiversity (Sobral-Souza et al. 2021b, Santos et al. 2020). Like any other deficit on biodiversity knowledge, the Wallacean Shortfall is dependent on the spatial scale on which it is assessed. At high spatial resolution for a wide spatial scale the knowledge of biodiversity can be considered complete as all grid cells would have enough occurrences to reach a high degree of completeness (Lobo et al. 2018). But at low spatial resolution, as here, we infer that there are many spatial gaps in the knowledge of the biodiversity of small mammal in the Atlantic Forest.
Currently, the biodiversity knowledge of small rodents in the Atlantic Forest is insufficient to understand how landscape modification affects the spatial species distribution on a large scale. The bias highlighted here demonstrated a biodiversity knowledge gap for small fragments, with little habitat (forest) coverage and far from roads and cities. Most of the Atlantic Forest area is represented by small fragments, with little forest cover, as they are composed of secondary forests or in forest recovery (Ribeiro et al. 2009). In addition, the well-sampled sites are spatially aggregated, not covering the entire range of accessibility and landscape conditions of the Atlantic Forest. Therefore, samplings of small rodents need to be directed to sites of difficult accessibility, away from cities, in small fragments and with little forest habitat cover in order to contribute to the increase of the inventory completeness through the landscape configuration spectrum mainly due to the dependence on responses that small rodents can present with the landscape human-induced modification.