493
and sampling efforts together with the data (https://www.gbif.org/darwin-core).
Full documentation of sampling events using adequate metadata standards not
only is critical for facilitating reuse of data by secondary users but also provides
important information for quantifying the associated uncertainty and eventually
applying correction techniques in subsequent steps. In practice, for a decade or
so, the critical importance of annotating data with standard metadata has guided
data management practices (e.g., in the context of long-term ecological research
networks; Michener et al. 2011). However, poor data practices that ignore the
annotation of metadata or that fail to adopt interoperable formats are still common for many biodiversity data sets, including those accessible through public
data archives of scientific journals (Roche et al. 2015). These deficiencies constitute a major bottleneck for building EBVs (Hugo et al. 2017).
2. Primary data aggregation. A second set of steps leads to the production of consolidated data products that typically conform to all or most of the following
characteristics: They contain consistent biological quantities expressed in the
same measurement unit; other relevant biological attributes have been checked
and harmonized (e.g., into a harmonized taxonomy or a consistent typology of
traits or of ecosystem types); spatial and temporal references are matched; and
data uncertainties have been quantified. Standardized observations need to be
checked at this stage using quality control (QC) mechanisms that are documented
transparently (Rüegg et al. 2014), for example, looking at outliers to ensure data
quality. Collation of data in support of user requirements will ideally be automated using virtual infrastructures that are able to map the different (standardized) data sets with metadata into fully interoperable formats (Hugo et al. 2017).
As detailed in Kissling et al. (2018b), an excellent example of this for the Species
Traits EBVs is the Global Plant Phenology Data Portal (https://www.plantphenology.org), a platform that integrates phenology observations from three different networks using disparate data frameworks (Stucky et al. 2018). Key for this
integration was the design of a new “Plant Phenology Ontology” that was able to
provide a semantic framework as a basis to overcome interoperability problems
produced by network-specific terminologies for data recording. Finally, data
integration needs to deal with, and report on, uncertainties resulting from errors
that may propagate throughout the different EBV production steps, including
uncertain geographic locations of in-situ data, heterogeneous sampling methods
and efforts (Proença et al. 2017), and measurement errors.
3. Model-based estimation. Final EBV products ideally provide continuous information in space and at different time periods so biodiversity change can be measured throughout the entire spatial domain. This is the case for EBVs that can be
directly estimated using algorithms applied to satellite RS imagery with complete area coverage. On the contrary, for many EBVs that are primarily estimated
from in-situ data, an additional level of integration is required to overcome the
sparsity of data. Biodiversity models provide this level of integration by combining the strengths of in-situ observations and state-of-the-art RS products based
on correlative or deductive approaches (Jetz et al. 2012; Ferrier et al. 2017). For
instance, species distribution models are often based on a correlative relationship
18 Essential Biodiversity Variables: Integrating In-Situ Observations and Remote…
and sampling efforts together with the data (https://www.gbif.org/darwin-core).
Full documentation of sampling events using adequate metadata standards not
only is critical for facilitating reuse of data by secondary users but also provides
important information for quantifying the associated uncertainty and eventually
applying correction techniques in subsequent steps. In practice, for a decade or
so, the critical importance of annotating data with standard metadata has guided
data management practices (e.g., in the context of long-term ecological research
networks; Michener et al. 2011). However, poor data practices that ignore the
annotation of metadata or that fail to adopt interoperable formats are still common for many biodiversity data sets, including those accessible through public
data archives of scientific journals (Roche et al. 2015). These deficiencies constitute a major bottleneck for building EBVs (Hugo et al. 2017).
2. Primary data aggregation. A second set of steps leads to the production of consolidated data products that typically conform to all or most of the following
characteristics: They contain consistent biological quantities expressed in the
same measurement unit; other relevant biological attributes have been checked
and harmonized (e.g., into a harmonized taxonomy or a consistent typology of
traits or of ecosystem types); spatial and temporal references are matched; and
data uncertainties have been quantified. Standardized observations need to be
checked at this stage using quality control (QC) mechanisms that are documented
transparently (Rüegg et al. 2014), for example, looking at outliers to ensure data
quality. Collation of data in support of user requirements will ideally be automated using virtual infrastructures that are able to map the different (standardized) data sets with metadata into fully interoperable formats (Hugo et al. 2017).
As detailed in Kissling et al. (2018b), an excellent example of this for the Species
Traits EBVs is the Global Plant Phenology Data Portal (https://www.plantphenology.org), a platform that integrates phenology observations from three different networks using disparate data frameworks (Stucky et al. 2018). Key for this
integration was the design of a new “Plant Phenology Ontology” that was able to
provide a semantic framework as a basis to overcome interoperability problems
produced by network-specific terminologies for data recording. Finally, data
integration needs to deal with, and report on, uncertainties resulting from errors
that may propagate throughout the different EBV production steps, including
uncertain geographic locations of in-situ data, heterogeneous sampling methods
and efforts (Proença et al. 2017), and measurement errors.
3. Model-based estimation. Final EBV products ideally provide continuous information in space and at different time periods so biodiversity change can be measured throughout the entire spatial domain. This is the case for EBVs that can be
directly estimated using algorithms applied to satellite RS imagery with complete area coverage. On the contrary, for many EBVs that are primarily estimated
from in-situ data, an additional level of integration is required to overcome the
sparsity of data. Biodiversity models provide this level of integration by combining the strengths of in-situ observations and state-of-the-art RS products based
on correlative or deductive approaches (Jetz et al. 2012; Ferrier et al. 2017). For
instance, species distribution models are often based on a correlative relationship
18 Essential Biodiversity Variables: Integrating In-Situ Observations and Remote…