494
between environmental variables and the probability of the occurrence of a species. These models are calibrated or trained using species occurrence and sometimes absence data as response variables and environmental variables as
predictors. The probability of occurrence of a species can be spatially interpolated between the observation points because environmental variables are available as continuous surfaces (i.e., wall-to-wall), which are themselves generated
from models using in-situ and EO data. In deductive habitat modeling, expertbased assessment of the habitat preference and environmental constraints of species is used to refine the potential species distribution. When habitat predictors
are available in high resolution, this makes it possible to go from coarse potential
species distributions to fine-grain species distributions because species also
respond more locally to habitat variables than to, for instance, climate (Triviño
et al. 2011; Martins et al. 2014). Other EBVs can also be projected with models
that integrate in-situ observations with RS data and other environmental data.
Community composition variables such as the beta diversity between two sites
can be projected from climate and other variables using generalized dissimilarity
models (Ferrier et al. 2007), while alpha and gamma diversity can be projected
from land-use using the countryside species-area model (Pereira and Borda-deÁgua 2013). Hence, environmental predictors derived from RS constitute the
backbone of higher-resolution EBV products that are consistent in space and
time. However, it is important to note that such model-based EBVs provide
information that is fundamentally different from the aggregated data sets
described in the preceding steps and that while it improves the spatial and temporal coverage of the data set, it also introduces additional uncertainties that
need to be documented.
Massive integration of biodiversity data based on the EBV framework and workflows requires implementation via interoperable informatics infrastructures (Hugo
et al. 2017). Projects aligned with the mission and concepts of EBVs, such as Map
of Life (www.mol.org) or the Biogeographic Infrastructure for Large-scaled
Biodiversity Indicators (BILBI) (Hoskins et al. 2018), already constitute a proof of
concept of the potential of virtual infrastructures for developing a biodiversitymodeling framework that delivers global information from multi-sourced EO data
integration. While the technological implementation of these infrastructures should
not constitute a major limitation, redoubled efforts are needed, first, on making the
large amounts of in-situ data being collected available and interoperable and, second, on developing and adapting biodiversity models that are able to ingest massive
and novel sources of data, both in-situ (e.g., eDNA data) and RS (e.g., imaging
spectroscopy).
N. Fernández et al.
between environmental variables and the probability of the occurrence of a species. These models are calibrated or trained using species occurrence and sometimes absence data as response variables and environmental variables as
predictors. The probability of occurrence of a species can be spatially interpolated between the observation points because environmental variables are available as continuous surfaces (i.e., wall-to-wall), which are themselves generated
from models using in-situ and EO data. In deductive habitat modeling, expertbased assessment of the habitat preference and environmental constraints of species is used to refine the potential species distribution. When habitat predictors
are available in high resolution, this makes it possible to go from coarse potential
species distributions to fine-grain species distributions because species also
respond more locally to habitat variables than to, for instance, climate (Triviño
et al. 2011; Martins et al. 2014). Other EBVs can also be projected with models
that integrate in-situ observations with RS data and other environmental data.
Community composition variables such as the beta diversity between two sites
can be projected from climate and other variables using generalized dissimilarity
models (Ferrier et al. 2007), while alpha and gamma diversity can be projected
from land-use using the countryside species-area model (Pereira and Borda-deÁgua 2013). Hence, environmental predictors derived from RS constitute the
backbone of higher-resolution EBV products that are consistent in space and
time. However, it is important to note that such model-based EBVs provide
information that is fundamentally different from the aggregated data sets
described in the preceding steps and that while it improves the spatial and temporal coverage of the data set, it also introduces additional uncertainties that
need to be documented.
Massive integration of biodiversity data based on the EBV framework and workflows requires implementation via interoperable informatics infrastructures (Hugo
et al. 2017). Projects aligned with the mission and concepts of EBVs, such as Map
of Life (www.mol.org) or the Biogeographic Infrastructure for Large-scaled
Biodiversity Indicators (BILBI) (Hoskins et al. 2018), already constitute a proof of
concept of the potential of virtual infrastructures for developing a biodiversitymodeling framework that delivers global information from multi-sourced EO data
integration. While the technological implementation of these infrastructures should
not constitute a major limitation, redoubled efforts are needed, first, on making the
large amounts of in-situ data being collected available and interoperable and, second, on developing and adapting biodiversity models that are able to ingest massive
and novel sources of data, both in-situ (e.g., eDNA data) and RS (e.g., imaging
spectroscopy).
N. Fernández et al.